Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.10/site-packages/numpy/_core/_add_newdocs.py: 100%

Shortcuts on this page

r m x   toggle line displays

j k   next/prev highlighted chunk

0   (zero) top of page

1   (one) first highlighted chunk

270 statements  

1""" 

2This is only meant to add docs to objects defined in C-extension modules. 

3The purpose is to allow easier editing of the docstrings without 

4requiring a re-compile. 

5 

6NOTE: Many of the methods of ndarray have corresponding functions. 

7 If you update these docstrings, please keep also the ones in 

8 _core/fromnumeric.py, matrixlib/defmatrix.py up-to-date. 

9 

10""" 

11 

12from numpy._core.function_base import add_newdoc 

13from numpy._core.overrides import get_array_function_like_doc 

14 

15 

16############################################################################### 

17# 

18# flatiter 

19# 

20# flatiter needs a toplevel description 

21# 

22############################################################################### 

23 

24add_newdoc('numpy._core', 'flatiter', 

25 """ 

26 Flat iterator object to iterate over arrays. 

27 

28 A `flatiter` iterator is returned by ``x.flat`` for any array `x`. 

29 It allows iterating over the array as if it were a 1-D array, 

30 either in a for-loop or by calling its `next` method. 

31 

32 Iteration is done in row-major, C-style order (the last 

33 index varying the fastest). The iterator can also be indexed using 

34 basic slicing or advanced indexing. 

35 

36 See Also 

37 -------- 

38 ndarray.flat : Return a flat iterator over an array. 

39 ndarray.flatten : Returns a flattened copy of an array. 

40 

41 Notes 

42 ----- 

43 A `flatiter` iterator can not be constructed directly from Python code 

44 by calling the `flatiter` constructor. 

45 

46 Examples 

47 -------- 

48 >>> import numpy as np 

49 >>> x = np.arange(6).reshape(2, 3) 

50 >>> fl = x.flat 

51 >>> type(fl) 

52 <class 'numpy.flatiter'> 

53 >>> for item in fl: 

54 ... print(item) 

55 ... 

56 0 

57 1 

58 2 

59 3 

60 4 

61 5 

62 

63 >>> fl[2:4] 

64 array([2, 3]) 

65 

66 """) 

67 

68# flatiter attributes 

69 

70add_newdoc('numpy._core', 'flatiter', ('base', 

71 """ 

72 A reference to the array that is iterated over. 

73 

74 Examples 

75 -------- 

76 >>> import numpy as np 

77 >>> x = np.arange(5) 

78 >>> fl = x.flat 

79 >>> fl.base is x 

80 True 

81 

82 """)) 

83 

84 

85add_newdoc('numpy._core', 'flatiter', ('coords', 

86 """ 

87 An N-dimensional tuple of current coordinates. 

88 

89 Examples 

90 -------- 

91 >>> import numpy as np 

92 >>> x = np.arange(6).reshape(2, 3) 

93 >>> fl = x.flat 

94 >>> fl.coords 

95 (0, 0) 

96 >>> next(fl) 

97 0 

98 >>> fl.coords 

99 (0, 1) 

100 

101 """)) 

102 

103 

104add_newdoc('numpy._core', 'flatiter', ('index', 

105 """ 

106 Current flat index into the array. 

107 

108 Examples 

109 -------- 

110 >>> import numpy as np 

111 >>> x = np.arange(6).reshape(2, 3) 

112 >>> fl = x.flat 

113 >>> fl.index 

114 0 

115 >>> next(fl) 

116 0 

117 >>> fl.index 

118 1 

119 

120 """)) 

121 

122# flatiter functions 

123 

124add_newdoc('numpy._core', 'flatiter', ('__array__', 

125 """__array__(type=None) Get array from iterator 

126 

127 """)) 

128 

129 

130add_newdoc('numpy._core', 'flatiter', ('copy', 

131 """ 

132 copy() 

133 

134 Get a copy of the iterator as a 1-D array. 

135 

136 Examples 

137 -------- 

138 >>> import numpy as np 

139 >>> x = np.arange(6).reshape(2, 3) 

140 >>> x 

141 array([[0, 1, 2], 

142 [3, 4, 5]]) 

143 >>> fl = x.flat 

144 >>> fl.copy() 

145 array([0, 1, 2, 3, 4, 5]) 

146 

147 """)) 

148 

149 

150############################################################################### 

151# 

152# nditer 

153# 

154############################################################################### 

155 

156add_newdoc('numpy._core', 'nditer', 

157 """ 

158 nditer(op, flags=None, op_flags=None, op_dtypes=None, order='K', 

159 casting='safe', op_axes=None, itershape=None, buffersize=0) 

160 

161 Efficient multi-dimensional iterator object to iterate over arrays. 

162 To get started using this object, see the 

163 :ref:`introductory guide to array iteration <arrays.nditer>`. 

164 

165 Parameters 

166 ---------- 

167 op : ndarray or sequence of array_like 

168 The array(s) to iterate over. 

169 

170 flags : sequence of str, optional 

171 Flags to control the behavior of the iterator. 

172 

173 * ``buffered`` enables buffering when required. 

174 * ``c_index`` causes a C-order index to be tracked. 

175 * ``f_index`` causes a Fortran-order index to be tracked. 

176 * ``multi_index`` causes a multi-index, or a tuple of indices 

177 with one per iteration dimension, to be tracked. 

178 * ``common_dtype`` causes all the operands to be converted to 

179 a common data type, with copying or buffering as necessary. 

180 * ``copy_if_overlap`` causes the iterator to determine if read 

181 operands have overlap with write operands, and make temporary 

182 copies as necessary to avoid overlap. False positives (needless 

183 copying) are possible in some cases. 

184 * ``delay_bufalloc`` delays allocation of the buffers until 

185 a reset() call is made. Allows ``allocate`` operands to 

186 be initialized before their values are copied into the buffers. 

187 * ``external_loop`` causes the ``values`` given to be 

188 one-dimensional arrays with multiple values instead of 

189 zero-dimensional arrays. 

190 * ``grow_inner`` allows the ``value`` array sizes to be made 

191 larger than the buffer size when both ``buffered`` and 

192 ``external_loop`` is used. 

193 * ``ranged`` allows the iterator to be restricted to a sub-range 

194 of the iterindex values. 

195 * ``refs_ok`` enables iteration of reference types, such as 

196 object arrays. 

197 * ``reduce_ok`` enables iteration of ``readwrite`` operands 

198 which are broadcasted, also known as reduction operands. 

199 * ``zerosize_ok`` allows `itersize` to be zero. 

200 op_flags : list of list of str, optional 

201 This is a list of flags for each operand. At minimum, one of 

202 ``readonly``, ``readwrite``, or ``writeonly`` must be specified. 

203 

204 * ``readonly`` indicates the operand will only be read from. 

205 * ``readwrite`` indicates the operand will be read from and written to. 

206 * ``writeonly`` indicates the operand will only be written to. 

207 * ``no_broadcast`` prevents the operand from being broadcasted. 

208 * ``contig`` forces the operand data to be contiguous. 

209 * ``aligned`` forces the operand data to be aligned. 

210 * ``nbo`` forces the operand data to be in native byte order. 

211 * ``copy`` allows a temporary read-only copy if required. 

212 * ``updateifcopy`` allows a temporary read-write copy if required. 

213 * ``allocate`` causes the array to be allocated if it is None 

214 in the ``op`` parameter. 

215 * ``no_subtype`` prevents an ``allocate`` operand from using a subtype. 

216 * ``arraymask`` indicates that this operand is the mask to use 

217 for selecting elements when writing to operands with the 

218 'writemasked' flag set. The iterator does not enforce this, 

219 but when writing from a buffer back to the array, it only 

220 copies those elements indicated by this mask. 

221 * ``writemasked`` indicates that only elements where the chosen 

222 ``arraymask`` operand is True will be written to. 

223 * ``overlap_assume_elementwise`` can be used to mark operands that are 

224 accessed only in the iterator order, to allow less conservative 

225 copying when ``copy_if_overlap`` is present. 

226 op_dtypes : dtype or tuple of dtype(s), optional 

227 The required data type(s) of the operands. If copying or buffering 

228 is enabled, the data will be converted to/from their original types. 

229 order : {'C', 'F', 'A', 'K'}, optional 

230 Controls the iteration order. 'C' means C order, 'F' means 

231 Fortran order, 'A' means 'F' order if all the arrays are Fortran 

232 contiguous, 'C' order otherwise, and 'K' means as close to the 

233 order the array elements appear in memory as possible. This also 

234 affects the element memory order of ``allocate`` operands, as they 

235 are allocated to be compatible with iteration order. 

236 Default is 'K'. 

237 casting : {'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional 

238 Controls what kind of data casting may occur when making a copy 

239 or buffering. Setting this to 'unsafe' is not recommended, 

240 as it can adversely affect accumulations. 

241 

242 * 'no' means the data types should not be cast at all. 

243 * 'equiv' means only byte-order changes are allowed. 

244 * 'safe' means only casts which can preserve values are allowed. 

245 * 'same_kind' means only safe casts or casts within a kind, 

246 like float64 to float32, are allowed. 

247 * 'unsafe' means any data conversions may be done. 

248 op_axes : list of list of ints, optional 

249 If provided, is a list of ints or None for each operands. 

250 The list of axes for an operand is a mapping from the dimensions 

251 of the iterator to the dimensions of the operand. A value of 

252 -1 can be placed for entries, causing that dimension to be 

253 treated as `newaxis`. 

254 itershape : tuple of ints, optional 

255 The desired shape of the iterator. This allows ``allocate`` operands 

256 with a dimension mapped by op_axes not corresponding to a dimension 

257 of a different operand to get a value not equal to 1 for that 

258 dimension. 

259 buffersize : int, optional 

260 When buffering is enabled, controls the size of the temporary 

261 buffers. Set to 0 for the default value. 

262 

263 Attributes 

264 ---------- 

265 dtypes : tuple of dtype(s) 

266 The data types of the values provided in `value`. This may be 

267 different from the operand data types if buffering is enabled. 

268 Valid only before the iterator is closed. 

269 finished : bool 

270 Whether the iteration over the operands is finished or not. 

271 has_delayed_bufalloc : bool 

272 If True, the iterator was created with the ``delay_bufalloc`` flag, 

273 and no reset() function was called on it yet. 

274 has_index : bool 

275 If True, the iterator was created with either the ``c_index`` or 

276 the ``f_index`` flag, and the property `index` can be used to 

277 retrieve it. 

278 has_multi_index : bool 

279 If True, the iterator was created with the ``multi_index`` flag, 

280 and the property `multi_index` can be used to retrieve it. 

281 index 

282 When the ``c_index`` or ``f_index`` flag was used, this property 

283 provides access to the index. Raises a ValueError if accessed 

284 and ``has_index`` is False. 

285 iterationneedsapi : bool 

286 Whether iteration requires access to the Python API, for example 

287 if one of the operands is an object array. 

288 iterindex : int 

289 An index which matches the order of iteration. 

290 itersize : int 

291 Size of the iterator. 

292 itviews 

293 Structured view(s) of `operands` in memory, matching the reordered 

294 and optimized iterator access pattern. Valid only before the iterator 

295 is closed. 

296 multi_index 

297 When the ``multi_index`` flag was used, this property 

298 provides access to the index. Raises a ValueError if accessed 

299 accessed and ``has_multi_index`` is False. 

300 ndim : int 

301 The dimensions of the iterator. 

302 nop : int 

303 The number of iterator operands. 

304 operands : tuple of operand(s) 

305 The array(s) to be iterated over. Valid only before the iterator is 

306 closed. 

307 shape : tuple of ints 

308 Shape tuple, the shape of the iterator. 

309 value 

310 Value of ``operands`` at current iteration. Normally, this is a 

311 tuple of array scalars, but if the flag ``external_loop`` is used, 

312 it is a tuple of one dimensional arrays. 

313 

314 Notes 

315 ----- 

316 `nditer` supersedes `flatiter`. The iterator implementation behind 

317 `nditer` is also exposed by the NumPy C API. 

318 

319 The Python exposure supplies two iteration interfaces, one which follows 

320 the Python iterator protocol, and another which mirrors the C-style 

321 do-while pattern. The native Python approach is better in most cases, but 

322 if you need the coordinates or index of an iterator, use the C-style pattern. 

323 

324 Examples 

325 -------- 

326 Here is how we might write an ``iter_add`` function, using the 

327 Python iterator protocol: 

328 

329 >>> import numpy as np 

330 

331 >>> def iter_add_py(x, y, out=None): 

332 ... addop = np.add 

333 ... it = np.nditer([x, y, out], [], 

334 ... [['readonly'], ['readonly'], ['writeonly','allocate']]) 

335 ... with it: 

336 ... for (a, b, c) in it: 

337 ... addop(a, b, out=c) 

338 ... return it.operands[2] 

339 

340 Here is the same function, but following the C-style pattern: 

341 

342 >>> def iter_add(x, y, out=None): 

343 ... addop = np.add 

344 ... it = np.nditer([x, y, out], [], 

345 ... [['readonly'], ['readonly'], ['writeonly','allocate']]) 

346 ... with it: 

347 ... while not it.finished: 

348 ... addop(it[0], it[1], out=it[2]) 

349 ... it.iternext() 

350 ... return it.operands[2] 

351 

352 Here is an example outer product function: 

353 

354 >>> def outer_it(x, y, out=None): 

355 ... mulop = np.multiply 

356 ... it = np.nditer([x, y, out], ['external_loop'], 

357 ... [['readonly'], ['readonly'], ['writeonly', 'allocate']], 

358 ... op_axes=[list(range(x.ndim)) + [-1] * y.ndim, 

359 ... [-1] * x.ndim + list(range(y.ndim)), 

360 ... None]) 

361 ... with it: 

362 ... for (a, b, c) in it: 

363 ... mulop(a, b, out=c) 

364 ... return it.operands[2] 

365 

366 >>> a = np.arange(2)+1 

367 >>> b = np.arange(3)+1 

368 >>> outer_it(a,b) 

369 array([[1, 2, 3], 

370 [2, 4, 6]]) 

371 

372 Here is an example function which operates like a "lambda" ufunc: 

373 

374 >>> def luf(lamdaexpr, *args, **kwargs): 

375 ... '''luf(lambdaexpr, op1, ..., opn, out=None, order='K', casting='safe', buffersize=0)''' 

376 ... nargs = len(args) 

377 ... op = (kwargs.get('out',None),) + args 

378 ... it = np.nditer(op, ['buffered','external_loop'], 

379 ... [['writeonly','allocate','no_broadcast']] + 

380 ... [['readonly','nbo','aligned']]*nargs, 

381 ... order=kwargs.get('order','K'), 

382 ... casting=kwargs.get('casting','safe'), 

383 ... buffersize=kwargs.get('buffersize',0)) 

384 ... while not it.finished: 

385 ... it[0] = lamdaexpr(*it[1:]) 

386 ... it.iternext() 

387 ... return it.operands[0] 

388 

389 >>> a = np.arange(5) 

390 >>> b = np.ones(5) 

391 >>> luf(lambda i,j:i*i + j/2, a, b) 

392 array([ 0.5, 1.5, 4.5, 9.5, 16.5]) 

393 

394 If operand flags ``"writeonly"`` or ``"readwrite"`` are used the 

395 operands may be views into the original data with the 

396 `WRITEBACKIFCOPY` flag. In this case `nditer` must be used as a 

397 context manager or the `nditer.close` method must be called before 

398 using the result. The temporary data will be written back to the 

399 original data when the :meth:`~object.__exit__` function is called 

400 but not before: 

401 

402 >>> a = np.arange(6, dtype='i4')[::-2] 

403 >>> with np.nditer(a, [], 

404 ... [['writeonly', 'updateifcopy']], 

405 ... casting='unsafe', 

406 ... op_dtypes=[np.dtype('f4')]) as i: 

407 ... x = i.operands[0] 

408 ... x[:] = [-1, -2, -3] 

409 ... # a still unchanged here 

410 >>> a, x 

411 (array([-1, -2, -3], dtype=int32), array([-1., -2., -3.], dtype=float32)) 

412 

413 It is important to note that once the iterator is exited, dangling 

414 references (like `x` in the example) may or may not share data with 

415 the original data `a`. If writeback semantics were active, i.e. if 

416 `x.base.flags.writebackifcopy` is `True`, then exiting the iterator 

417 will sever the connection between `x` and `a`, writing to `x` will 

418 no longer write to `a`. If writeback semantics are not active, then 

419 `x.data` will still point at some part of `a.data`, and writing to 

420 one will affect the other. 

421 

422 Context management and the `close` method appeared in version 1.15.0. 

423 

424 """) 

425 

426# nditer methods 

427 

428add_newdoc('numpy._core', 'nditer', ('copy', 

429 """ 

430 copy() 

431 

432 Get a copy of the iterator in its current state. 

433 

434 Examples 

435 -------- 

436 >>> import numpy as np 

437 >>> x = np.arange(10) 

438 >>> y = x + 1 

439 >>> it = np.nditer([x, y]) 

440 >>> next(it) 

441 (array(0), array(1)) 

442 >>> it2 = it.copy() 

443 >>> next(it2) 

444 (array(1), array(2)) 

445 

446 """)) 

447 

448add_newdoc('numpy._core', 'nditer', ('operands', 

449 """ 

450 operands[`Slice`] 

451 

452 The array(s) to be iterated over. Valid only before the iterator is closed. 

453 """)) 

454 

455add_newdoc('numpy._core', 'nditer', ('debug_print', 

456 """ 

457 debug_print() 

458 

459 Print the current state of the `nditer` instance and debug info to stdout. 

460 

461 """)) 

462 

463add_newdoc('numpy._core', 'nditer', ('enable_external_loop', 

464 """ 

465 enable_external_loop() 

466 

467 When the "external_loop" was not used during construction, but 

468 is desired, this modifies the iterator to behave as if the flag 

469 was specified. 

470 

471 """)) 

472 

473add_newdoc('numpy._core', 'nditer', ('iternext', 

474 """ 

475 iternext() 

476 

477 Check whether iterations are left, and perform a single internal iteration 

478 without returning the result. Used in the C-style pattern do-while 

479 pattern. For an example, see `nditer`. 

480 

481 Returns 

482 ------- 

483 iternext : bool 

484 Whether or not there are iterations left. 

485 

486 """)) 

487 

488add_newdoc('numpy._core', 'nditer', ('remove_axis', 

489 """ 

490 remove_axis(i, /) 

491 

492 Removes axis `i` from the iterator. Requires that the flag "multi_index" 

493 be enabled. 

494 

495 """)) 

496 

497add_newdoc('numpy._core', 'nditer', ('remove_multi_index', 

498 """ 

499 remove_multi_index() 

500 

501 When the "multi_index" flag was specified, this removes it, allowing 

502 the internal iteration structure to be optimized further. 

503 

504 """)) 

505 

506add_newdoc('numpy._core', 'nditer', ('reset', 

507 """ 

508 reset() 

509 

510 Reset the iterator to its initial state. 

511 

512 """)) 

513 

514add_newdoc('numpy._core', 'nested_iters', 

515 """ 

516 nested_iters(op, axes, flags=None, op_flags=None, op_dtypes=None, \ 

517 order="K", casting="safe", buffersize=0) 

518 

519 Create nditers for use in nested loops 

520 

521 Create a tuple of `nditer` objects which iterate in nested loops over 

522 different axes of the op argument. The first iterator is used in the 

523 outermost loop, the last in the innermost loop. Advancing one will change 

524 the subsequent iterators to point at its new element. 

525 

526 Parameters 

527 ---------- 

528 op : ndarray or sequence of array_like 

529 The array(s) to iterate over. 

530 

531 axes : list of list of int 

532 Each item is used as an "op_axes" argument to an nditer 

533 

534 flags, op_flags, op_dtypes, order, casting, buffersize (optional) 

535 See `nditer` parameters of the same name 

536 

537 Returns 

538 ------- 

539 iters : tuple of nditer 

540 An nditer for each item in `axes`, outermost first 

541 

542 See Also 

543 -------- 

544 nditer 

545 

546 Examples 

547 -------- 

548 

549 Basic usage. Note how y is the "flattened" version of 

550 [a[:, 0, :], a[:, 1, 0], a[:, 2, :]] since we specified 

551 the first iter's axes as [1] 

552 

553 >>> import numpy as np 

554 >>> a = np.arange(12).reshape(2, 3, 2) 

555 >>> i, j = np.nested_iters(a, [[1], [0, 2]], flags=["multi_index"]) 

556 >>> for x in i: 

557 ... print(i.multi_index) 

558 ... for y in j: 

559 ... print('', j.multi_index, y) 

560 (0,) 

561 (0, 0) 0 

562 (0, 1) 1 

563 (1, 0) 6 

564 (1, 1) 7 

565 (1,) 

566 (0, 0) 2 

567 (0, 1) 3 

568 (1, 0) 8 

569 (1, 1) 9 

570 (2,) 

571 (0, 0) 4 

572 (0, 1) 5 

573 (1, 0) 10 

574 (1, 1) 11 

575 

576 """) 

577 

578add_newdoc('numpy._core', 'nditer', ('close', 

579 """ 

580 close() 

581 

582 Resolve all writeback semantics in writeable operands. 

583 

584 See Also 

585 -------- 

586 

587 :ref:`nditer-context-manager` 

588 

589 """)) 

590 

591 

592############################################################################### 

593# 

594# broadcast 

595# 

596############################################################################### 

597 

598add_newdoc('numpy._core', 'broadcast', 

599 """ 

600 Produce an object that mimics broadcasting. 

601 

602 Parameters 

603 ---------- 

604 in1, in2, ... : array_like 

605 Input parameters. 

606 

607 Returns 

608 ------- 

609 b : broadcast object 

610 Broadcast the input parameters against one another, and 

611 return an object that encapsulates the result. 

612 Amongst others, it has ``shape`` and ``nd`` properties, and 

613 may be used as an iterator. 

614 

615 See Also 

616 -------- 

617 broadcast_arrays 

618 broadcast_to 

619 broadcast_shapes 

620 

621 Examples 

622 -------- 

623 

624 Manually adding two vectors, using broadcasting: 

625 

626 >>> import numpy as np 

627 >>> x = np.array([[1], [2], [3]]) 

628 >>> y = np.array([4, 5, 6]) 

629 >>> b = np.broadcast(x, y) 

630 

631 >>> out = np.empty(b.shape) 

632 >>> out.flat = [u+v for (u,v) in b] 

633 >>> out 

634 array([[5., 6., 7.], 

635 [6., 7., 8.], 

636 [7., 8., 9.]]) 

637 

638 Compare against built-in broadcasting: 

639 

640 >>> x + y 

641 array([[5, 6, 7], 

642 [6, 7, 8], 

643 [7, 8, 9]]) 

644 

645 """) 

646 

647# attributes 

648 

649add_newdoc('numpy._core', 'broadcast', ('index', 

650 """ 

651 current index in broadcasted result 

652 

653 Examples 

654 -------- 

655 

656 >>> import numpy as np 

657 >>> x = np.array([[1], [2], [3]]) 

658 >>> y = np.array([4, 5, 6]) 

659 >>> b = np.broadcast(x, y) 

660 >>> b.index 

661 0 

662 >>> next(b), next(b), next(b) 

663 ((1, 4), (1, 5), (1, 6)) 

664 >>> b.index 

665 3 

666 

667 """)) 

668 

669add_newdoc('numpy._core', 'broadcast', ('iters', 

670 """ 

671 tuple of iterators along ``self``'s "components." 

672 

673 Returns a tuple of `numpy.flatiter` objects, one for each "component" 

674 of ``self``. 

675 

676 See Also 

677 -------- 

678 numpy.flatiter 

679 

680 Examples 

681 -------- 

682 

683 >>> import numpy as np 

684 >>> x = np.array([1, 2, 3]) 

685 >>> y = np.array([[4], [5], [6]]) 

686 >>> b = np.broadcast(x, y) 

687 >>> row, col = b.iters 

688 >>> next(row), next(col) 

689 (1, 4) 

690 

691 """)) 

692 

693add_newdoc('numpy._core', 'broadcast', ('ndim', 

694 """ 

695 Number of dimensions of broadcasted result. Alias for `nd`. 

696 

697 Examples 

698 -------- 

699 >>> import numpy as np 

700 >>> x = np.array([1, 2, 3]) 

701 >>> y = np.array([[4], [5], [6]]) 

702 >>> b = np.broadcast(x, y) 

703 >>> b.ndim 

704 2 

705 

706 """)) 

707 

708add_newdoc('numpy._core', 'broadcast', ('nd', 

709 """ 

710 Number of dimensions of broadcasted result. For code intended for NumPy 

711 1.12.0 and later the more consistent `ndim` is preferred. 

712 

713 Examples 

714 -------- 

715 >>> import numpy as np 

716 >>> x = np.array([1, 2, 3]) 

717 >>> y = np.array([[4], [5], [6]]) 

718 >>> b = np.broadcast(x, y) 

719 >>> b.nd 

720 2 

721 

722 """)) 

723 

724add_newdoc('numpy._core', 'broadcast', ('numiter', 

725 """ 

726 Number of iterators possessed by the broadcasted result. 

727 

728 Examples 

729 -------- 

730 >>> import numpy as np 

731 >>> x = np.array([1, 2, 3]) 

732 >>> y = np.array([[4], [5], [6]]) 

733 >>> b = np.broadcast(x, y) 

734 >>> b.numiter 

735 2 

736 

737 """)) 

738 

739add_newdoc('numpy._core', 'broadcast', ('shape', 

740 """ 

741 Shape of broadcasted result. 

742 

743 Examples 

744 -------- 

745 >>> import numpy as np 

746 >>> x = np.array([1, 2, 3]) 

747 >>> y = np.array([[4], [5], [6]]) 

748 >>> b = np.broadcast(x, y) 

749 >>> b.shape 

750 (3, 3) 

751 

752 """)) 

753 

754add_newdoc('numpy._core', 'broadcast', ('size', 

755 """ 

756 Total size of broadcasted result. 

757 

758 Examples 

759 -------- 

760 >>> import numpy as np 

761 >>> x = np.array([1, 2, 3]) 

762 >>> y = np.array([[4], [5], [6]]) 

763 >>> b = np.broadcast(x, y) 

764 >>> b.size 

765 9 

766 

767 """)) 

768 

769add_newdoc('numpy._core', 'broadcast', ('reset', 

770 """ 

771 reset() 

772 

773 Reset the broadcasted result's iterator(s). 

774 

775 Parameters 

776 ---------- 

777 None 

778 

779 Returns 

780 ------- 

781 None 

782 

783 Examples 

784 -------- 

785 >>> import numpy as np 

786 >>> x = np.array([1, 2, 3]) 

787 >>> y = np.array([[4], [5], [6]]) 

788 >>> b = np.broadcast(x, y) 

789 >>> b.index 

790 0 

791 >>> next(b), next(b), next(b) 

792 ((1, 4), (2, 4), (3, 4)) 

793 >>> b.index 

794 3 

795 >>> b.reset() 

796 >>> b.index 

797 0 

798 

799 """)) 

800 

801############################################################################### 

802# 

803# numpy functions 

804# 

805############################################################################### 

806 

807add_newdoc('numpy._core.multiarray', 'array', 

808 """ 

809 array(object, dtype=None, *, copy=True, order='K', subok=False, ndmin=0, 

810 like=None) 

811 

812 Create an array. 

813 

814 Parameters 

815 ---------- 

816 object : array_like 

817 An array, any object exposing the array interface, an object whose 

818 ``__array__`` method returns an array, or any (nested) sequence. 

819 If object is a scalar, a 0-dimensional array containing object is 

820 returned. 

821 dtype : data-type, optional 

822 The desired data-type for the array. If not given, NumPy will try to use 

823 a default ``dtype`` that can represent the values (by applying promotion 

824 rules when necessary.) 

825 copy : bool, optional 

826 If ``True`` (default), then the array data is copied. If ``None``, 

827 a copy will only be made if ``__array__`` returns a copy, if obj is 

828 a nested sequence, or if a copy is needed to satisfy any of the other 

829 requirements (``dtype``, ``order``, etc.). Note that any copy of 

830 the data is shallow, i.e., for arrays with object dtype, the new 

831 array will point to the same objects. See Examples for `ndarray.copy`. 

832 For ``False`` it raises a ``ValueError`` if a copy cannot be avoided. 

833 Default: ``True``. 

834 order : {'K', 'A', 'C', 'F'}, optional 

835 Specify the memory layout of the array. If object is not an array, the 

836 newly created array will be in C order (row major) unless 'F' is 

837 specified, in which case it will be in Fortran order (column major). 

838 If object is an array the following holds. 

839 

840 ===== ========= =================================================== 

841 order no copy copy=True 

842 ===== ========= =================================================== 

843 'K' unchanged F & C order preserved, otherwise most similar order 

844 'A' unchanged F order if input is F and not C, otherwise C order 

845 'C' C order C order 

846 'F' F order F order 

847 ===== ========= =================================================== 

848 

849 When ``copy=None`` and a copy is made for other reasons, the result is 

850 the same as if ``copy=True``, with some exceptions for 'A', see the 

851 Notes section. The default order is 'K'. 

852 subok : bool, optional 

853 If True, then sub-classes will be passed-through, otherwise 

854 the returned array will be forced to be a base-class array (default). 

855 ndmin : int, optional 

856 Specifies the minimum number of dimensions that the resulting 

857 array should have. Ones will be prepended to the shape as 

858 needed to meet this requirement. 

859 ${ARRAY_FUNCTION_LIKE} 

860 

861 .. versionadded:: 1.20.0 

862 

863 Returns 

864 ------- 

865 out : ndarray 

866 An array object satisfying the specified requirements. 

867 

868 See Also 

869 -------- 

870 empty_like : Return an empty array with shape and type of input. 

871 ones_like : Return an array of ones with shape and type of input. 

872 zeros_like : Return an array of zeros with shape and type of input. 

873 full_like : Return a new array with shape of input filled with value. 

874 empty : Return a new uninitialized array. 

875 ones : Return a new array setting values to one. 

876 zeros : Return a new array setting values to zero. 

877 full : Return a new array of given shape filled with value. 

878 copy: Return an array copy of the given object. 

879 

880 

881 Notes 

882 ----- 

883 When order is 'A' and ``object`` is an array in neither 'C' nor 'F' order, 

884 and a copy is forced by a change in dtype, then the order of the result is 

885 not necessarily 'C' as expected. This is likely a bug. 

886 

887 Examples 

888 -------- 

889 >>> import numpy as np 

890 >>> np.array([1, 2, 3]) 

891 array([1, 2, 3]) 

892 

893 Upcasting: 

894 

895 >>> np.array([1, 2, 3.0]) 

896 array([ 1., 2., 3.]) 

897 

898 More than one dimension: 

899 

900 >>> np.array([[1, 2], [3, 4]]) 

901 array([[1, 2], 

902 [3, 4]]) 

903 

904 Minimum dimensions 2: 

905 

906 >>> np.array([1, 2, 3], ndmin=2) 

907 array([[1, 2, 3]]) 

908 

909 Type provided: 

910 

911 >>> np.array([1, 2, 3], dtype=complex) 

912 array([ 1.+0.j, 2.+0.j, 3.+0.j]) 

913 

914 Data-type consisting of more than one element: 

915 

916 >>> x = np.array([(1,2),(3,4)],dtype=[('a','<i4'),('b','<i4')]) 

917 >>> x['a'] 

918 array([1, 3], dtype=int32) 

919 

920 Creating an array from sub-classes: 

921 

922 >>> np.array(np.asmatrix('1 2; 3 4')) 

923 array([[1, 2], 

924 [3, 4]]) 

925 

926 >>> np.array(np.asmatrix('1 2; 3 4'), subok=True) 

927 matrix([[1, 2], 

928 [3, 4]]) 

929 

930 """) 

931 

932add_newdoc('numpy._core.multiarray', 'asarray', 

933 """ 

934 asarray(a, dtype=None, order=None, *, device=None, copy=None, like=None) 

935 

936 Convert the input to an array. 

937 

938 Parameters 

939 ---------- 

940 a : array_like 

941 Input data, in any form that can be converted to an array. This 

942 includes lists, lists of tuples, tuples, tuples of tuples, tuples 

943 of lists and ndarrays. 

944 dtype : data-type, optional 

945 By default, the data-type is inferred from the input data. 

946 order : {'C', 'F', 'A', 'K'}, optional 

947 Memory layout. 'A' and 'K' depend on the order of input array a. 

948 'C' row-major (C-style), 

949 'F' column-major (Fortran-style) memory representation. 

950 'A' (any) means 'F' if `a` is Fortran contiguous, 'C' otherwise 

951 'K' (keep) preserve input order 

952 Defaults to 'K'. 

953 device : str, optional 

954 The device on which to place the created array. Default: ``None``. 

955 For Array-API interoperability only, so must be ``"cpu"`` if passed. 

956 

957 .. versionadded:: 2.0.0 

958 copy : bool, optional 

959 If ``True``, then the object is copied. If ``None`` then the object is 

960 copied only if needed, i.e. if ``__array__`` returns a copy, if obj 

961 is a nested sequence, or if a copy is needed to satisfy any of 

962 the other requirements (``dtype``, ``order``, etc.). 

963 For ``False`` it raises a ``ValueError`` if a copy cannot be avoided. 

964 Default: ``None``. 

965 

966 .. versionadded:: 2.0.0 

967 ${ARRAY_FUNCTION_LIKE} 

968 

969 .. versionadded:: 1.20.0 

970 

971 Returns 

972 ------- 

973 out : ndarray 

974 Array interpretation of ``a``. No copy is performed if the input 

975 is already an ndarray with matching dtype and order. If ``a`` is a 

976 subclass of ndarray, a base class ndarray is returned. 

977 

978 See Also 

979 -------- 

980 asanyarray : Similar function which passes through subclasses. 

981 ascontiguousarray : Convert input to a contiguous array. 

982 asfortranarray : Convert input to an ndarray with column-major 

983 memory order. 

984 asarray_chkfinite : Similar function which checks input for NaNs and Infs. 

985 fromiter : Create an array from an iterator. 

986 fromfunction : Construct an array by executing a function on grid 

987 positions. 

988 

989 Examples 

990 -------- 

991 Convert a list into an array: 

992 

993 >>> a = [1, 2] 

994 >>> import numpy as np 

995 >>> np.asarray(a) 

996 array([1, 2]) 

997 

998 Existing arrays are not copied: 

999 

1000 >>> a = np.array([1, 2]) 

1001 >>> np.asarray(a) is a 

1002 True 

1003 

1004 If `dtype` is set, array is copied only if dtype does not match: 

1005 

1006 >>> a = np.array([1, 2], dtype=np.float32) 

1007 >>> np.shares_memory(np.asarray(a, dtype=np.float32), a) 

1008 True 

1009 >>> np.shares_memory(np.asarray(a, dtype=np.float64), a) 

1010 False 

1011 

1012 Contrary to `asanyarray`, ndarray subclasses are not passed through: 

1013 

1014 >>> issubclass(np.recarray, np.ndarray) 

1015 True 

1016 >>> a = np.array([(1., 2), (3., 4)], dtype='f4,i4').view(np.recarray) 

1017 >>> np.asarray(a) is a 

1018 False 

1019 >>> np.asanyarray(a) is a 

1020 True 

1021 

1022 """) 

1023 

1024add_newdoc('numpy._core.multiarray', 'asanyarray', 

1025 """ 

1026 asanyarray(a, dtype=None, order=None, *, device=None, copy=None, like=None) 

1027 

1028 Convert the input to an ndarray, but pass ndarray subclasses through. 

1029 

1030 Parameters 

1031 ---------- 

1032 a : array_like 

1033 Input data, in any form that can be converted to an array. This 

1034 includes scalars, lists, lists of tuples, tuples, tuples of tuples, 

1035 tuples of lists, and ndarrays. 

1036 dtype : data-type, optional 

1037 By default, the data-type is inferred from the input data. 

1038 order : {'C', 'F', 'A', 'K'}, optional 

1039 Memory layout. 'A' and 'K' depend on the order of input array a. 

1040 'C' row-major (C-style), 

1041 'F' column-major (Fortran-style) memory representation. 

1042 'A' (any) means 'F' if `a` is Fortran contiguous, 'C' otherwise 

1043 'K' (keep) preserve input order 

1044 Defaults to 'C'. 

1045 device : str, optional 

1046 The device on which to place the created array. Default: ``None``. 

1047 For Array-API interoperability only, so must be ``"cpu"`` if passed. 

1048 

1049 .. versionadded:: 2.1.0 

1050 

1051 copy : bool, optional 

1052 If ``True``, then the object is copied. If ``None`` then the object is 

1053 copied only if needed, i.e. if ``__array__`` returns a copy, if obj 

1054 is a nested sequence, or if a copy is needed to satisfy any of 

1055 the other requirements (``dtype``, ``order``, etc.). 

1056 For ``False`` it raises a ``ValueError`` if a copy cannot be avoided. 

1057 Default: ``None``. 

1058 

1059 .. versionadded:: 2.1.0 

1060 

1061 ${ARRAY_FUNCTION_LIKE} 

1062 

1063 .. versionadded:: 1.20.0 

1064 

1065 Returns 

1066 ------- 

1067 out : ndarray or an ndarray subclass 

1068 Array interpretation of `a`. If `a` is an ndarray or a subclass 

1069 of ndarray, it is returned as-is and no copy is performed. 

1070 

1071 See Also 

1072 -------- 

1073 asarray : Similar function which always returns ndarrays. 

1074 ascontiguousarray : Convert input to a contiguous array. 

1075 asfortranarray : Convert input to an ndarray with column-major 

1076 memory order. 

1077 asarray_chkfinite : Similar function which checks input for NaNs and 

1078 Infs. 

1079 fromiter : Create an array from an iterator. 

1080 fromfunction : Construct an array by executing a function on grid 

1081 positions. 

1082 

1083 Examples 

1084 -------- 

1085 Convert a list into an array: 

1086 

1087 >>> a = [1, 2] 

1088 >>> import numpy as np 

1089 >>> np.asanyarray(a) 

1090 array([1, 2]) 

1091 

1092 Instances of `ndarray` subclasses are passed through as-is: 

1093 

1094 >>> a = np.array([(1., 2), (3., 4)], dtype='f4,i4').view(np.recarray) 

1095 >>> np.asanyarray(a) is a 

1096 True 

1097 

1098 """) 

1099 

1100add_newdoc('numpy._core.multiarray', 'ascontiguousarray', 

1101 """ 

1102 ascontiguousarray(a, dtype=None, *, like=None) 

1103 

1104 Return a contiguous array (ndim >= 1) in memory (C order). 

1105 

1106 Parameters 

1107 ---------- 

1108 a : array_like 

1109 Input array. 

1110 dtype : str or dtype object, optional 

1111 Data-type of returned array. 

1112 ${ARRAY_FUNCTION_LIKE} 

1113 

1114 .. versionadded:: 1.20.0 

1115 

1116 Returns 

1117 ------- 

1118 out : ndarray 

1119 Contiguous array of same shape and content as `a`, with type `dtype` 

1120 if specified. 

1121 

1122 See Also 

1123 -------- 

1124 asfortranarray : Convert input to an ndarray with column-major 

1125 memory order. 

1126 require : Return an ndarray that satisfies requirements. 

1127 ndarray.flags : Information about the memory layout of the array. 

1128 

1129 Examples 

1130 -------- 

1131 Starting with a Fortran-contiguous array: 

1132 

1133 >>> import numpy as np 

1134 >>> x = np.ones((2, 3), order='F') 

1135 >>> x.flags['F_CONTIGUOUS'] 

1136 True 

1137 

1138 Calling ``ascontiguousarray`` makes a C-contiguous copy: 

1139 

1140 >>> y = np.ascontiguousarray(x) 

1141 >>> y.flags['C_CONTIGUOUS'] 

1142 True 

1143 >>> np.may_share_memory(x, y) 

1144 False 

1145 

1146 Now, starting with a C-contiguous array: 

1147 

1148 >>> x = np.ones((2, 3), order='C') 

1149 >>> x.flags['C_CONTIGUOUS'] 

1150 True 

1151 

1152 Then, calling ``ascontiguousarray`` returns the same object: 

1153 

1154 >>> y = np.ascontiguousarray(x) 

1155 >>> x is y 

1156 True 

1157 

1158 Note: This function returns an array with at least one-dimension (1-d) 

1159 so it will not preserve 0-d arrays. 

1160 

1161 """) 

1162 

1163add_newdoc('numpy._core.multiarray', 'asfortranarray', 

1164 """ 

1165 asfortranarray(a, dtype=None, *, like=None) 

1166 

1167 Return an array (ndim >= 1) laid out in Fortran order in memory. 

1168 

1169 Parameters 

1170 ---------- 

1171 a : array_like 

1172 Input array. 

1173 dtype : str or dtype object, optional 

1174 By default, the data-type is inferred from the input data. 

1175 ${ARRAY_FUNCTION_LIKE} 

1176 

1177 .. versionadded:: 1.20.0 

1178 

1179 Returns 

1180 ------- 

1181 out : ndarray 

1182 The input `a` in Fortran, or column-major, order. 

1183 

1184 See Also 

1185 -------- 

1186 ascontiguousarray : Convert input to a contiguous (C order) array. 

1187 asanyarray : Convert input to an ndarray with either row or 

1188 column-major memory order. 

1189 require : Return an ndarray that satisfies requirements. 

1190 ndarray.flags : Information about the memory layout of the array. 

1191 

1192 Examples 

1193 -------- 

1194 Starting with a C-contiguous array: 

1195 

1196 >>> import numpy as np 

1197 >>> x = np.ones((2, 3), order='C') 

1198 >>> x.flags['C_CONTIGUOUS'] 

1199 True 

1200 

1201 Calling ``asfortranarray`` makes a Fortran-contiguous copy: 

1202 

1203 >>> y = np.asfortranarray(x) 

1204 >>> y.flags['F_CONTIGUOUS'] 

1205 True 

1206 >>> np.may_share_memory(x, y) 

1207 False 

1208 

1209 Now, starting with a Fortran-contiguous array: 

1210 

1211 >>> x = np.ones((2, 3), order='F') 

1212 >>> x.flags['F_CONTIGUOUS'] 

1213 True 

1214 

1215 Then, calling ``asfortranarray`` returns the same object: 

1216 

1217 >>> y = np.asfortranarray(x) 

1218 >>> x is y 

1219 True 

1220 

1221 Note: This function returns an array with at least one-dimension (1-d) 

1222 so it will not preserve 0-d arrays. 

1223 

1224 """) 

1225 

1226add_newdoc('numpy._core.multiarray', 'empty', 

1227 """ 

1228 empty(shape, dtype=float, order='C', *, device=None, like=None) 

1229 

1230 Return a new array of given shape and type, without initializing entries. 

1231 

1232 Parameters 

1233 ---------- 

1234 shape : int or tuple of int 

1235 Shape of the empty array, e.g., ``(2, 3)`` or ``2``. 

1236 dtype : data-type, optional 

1237 Desired output data-type for the array, e.g, `numpy.int8`. Default is 

1238 `numpy.float64`. 

1239 order : {'C', 'F'}, optional, default: 'C' 

1240 Whether to store multi-dimensional data in row-major 

1241 (C-style) or column-major (Fortran-style) order in 

1242 memory. 

1243 device : str, optional 

1244 The device on which to place the created array. Default: ``None``. 

1245 For Array-API interoperability only, so must be ``"cpu"`` if passed. 

1246 

1247 .. versionadded:: 2.0.0 

1248 ${ARRAY_FUNCTION_LIKE} 

1249 

1250 .. versionadded:: 1.20.0 

1251 

1252 Returns 

1253 ------- 

1254 out : ndarray 

1255 Array of uninitialized (arbitrary) data of the given shape, dtype, and 

1256 order. Object arrays will be initialized to None. 

1257 

1258 See Also 

1259 -------- 

1260 empty_like : Return an empty array with shape and type of input. 

1261 ones : Return a new array setting values to one. 

1262 zeros : Return a new array setting values to zero. 

1263 full : Return a new array of given shape filled with value. 

1264 

1265 Notes 

1266 ----- 

1267 Unlike other array creation functions (e.g. `zeros`, `ones`, `full`), 

1268 `empty` does not initialize the values of the array, and may therefore be 

1269 marginally faster. However, the values stored in the newly allocated array 

1270 are arbitrary. For reproducible behavior, be sure to set each element of 

1271 the array before reading. 

1272 

1273 Examples 

1274 -------- 

1275 >>> import numpy as np 

1276 >>> np.empty([2, 2]) 

1277 array([[ -9.74499359e+001, 6.69583040e-309], 

1278 [ 2.13182611e-314, 3.06959433e-309]]) #uninitialized 

1279 

1280 >>> np.empty([2, 2], dtype=int) 

1281 array([[-1073741821, -1067949133], 

1282 [ 496041986, 19249760]]) #uninitialized 

1283 

1284 """) 

1285 

1286add_newdoc('numpy._core.multiarray', 'scalar', 

1287 """ 

1288 scalar(dtype, obj) 

1289 

1290 Return a new scalar array of the given type initialized with obj. 

1291 

1292 This function is meant mainly for pickle support. `dtype` must be a 

1293 valid data-type descriptor. If `dtype` corresponds to an object 

1294 descriptor, then `obj` can be any object, otherwise `obj` must be a 

1295 string. If `obj` is not given, it will be interpreted as None for object 

1296 type and as zeros for all other types. 

1297 

1298 """) 

1299 

1300add_newdoc('numpy._core.multiarray', 'zeros', 

1301 """ 

1302 zeros(shape, dtype=float, order='C', *, like=None) 

1303 

1304 Return a new array of given shape and type, filled with zeros. 

1305 

1306 Parameters 

1307 ---------- 

1308 shape : int or tuple of ints 

1309 Shape of the new array, e.g., ``(2, 3)`` or ``2``. 

1310 dtype : data-type, optional 

1311 The desired data-type for the array, e.g., `numpy.int8`. Default is 

1312 `numpy.float64`. 

1313 order : {'C', 'F'}, optional, default: 'C' 

1314 Whether to store multi-dimensional data in row-major 

1315 (C-style) or column-major (Fortran-style) order in 

1316 memory. 

1317 ${ARRAY_FUNCTION_LIKE} 

1318 

1319 .. versionadded:: 1.20.0 

1320 

1321 Returns 

1322 ------- 

1323 out : ndarray 

1324 Array of zeros with the given shape, dtype, and order. 

1325 

1326 See Also 

1327 -------- 

1328 zeros_like : Return an array of zeros with shape and type of input. 

1329 empty : Return a new uninitialized array. 

1330 ones : Return a new array setting values to one. 

1331 full : Return a new array of given shape filled with value. 

1332 

1333 Examples 

1334 -------- 

1335 >>> import numpy as np 

1336 >>> np.zeros(5) 

1337 array([ 0., 0., 0., 0., 0.]) 

1338 

1339 >>> np.zeros((5,), dtype=int) 

1340 array([0, 0, 0, 0, 0]) 

1341 

1342 >>> np.zeros((2, 1)) 

1343 array([[ 0.], 

1344 [ 0.]]) 

1345 

1346 >>> s = (2,2) 

1347 >>> np.zeros(s) 

1348 array([[ 0., 0.], 

1349 [ 0., 0.]]) 

1350 

1351 >>> np.zeros((2,), dtype=[('x', 'i4'), ('y', 'i4')]) # custom dtype 

1352 array([(0, 0), (0, 0)], 

1353 dtype=[('x', '<i4'), ('y', '<i4')]) 

1354 

1355 """) 

1356 

1357add_newdoc('numpy._core.multiarray', 'set_typeDict', 

1358 """set_typeDict(dict) 

1359 

1360 Set the internal dictionary that can look up an array type using a 

1361 registered code. 

1362 

1363 """) 

1364 

1365add_newdoc('numpy._core.multiarray', 'fromstring', 

1366 """ 

1367 fromstring(string, dtype=float, count=-1, *, sep, like=None) 

1368 

1369 A new 1-D array initialized from text data in a string. 

1370 

1371 Parameters 

1372 ---------- 

1373 string : str 

1374 A string containing the data. 

1375 dtype : data-type, optional 

1376 The data type of the array; default: float. For binary input data, 

1377 the data must be in exactly this format. Most builtin numeric types are 

1378 supported and extension types may be supported. 

1379 count : int, optional 

1380 Read this number of `dtype` elements from the data. If this is 

1381 negative (the default), the count will be determined from the 

1382 length of the data. 

1383 sep : str, optional 

1384 The string separating numbers in the data; extra whitespace between 

1385 elements is also ignored. 

1386 

1387 .. deprecated:: 1.14 

1388 Passing ``sep=''``, the default, is deprecated since it will 

1389 trigger the deprecated binary mode of this function. This mode 

1390 interprets `string` as binary bytes, rather than ASCII text with 

1391 decimal numbers, an operation which is better spelt 

1392 ``frombuffer(string, dtype, count)``. If `string` contains unicode 

1393 text, the binary mode of `fromstring` will first encode it into 

1394 bytes using utf-8, which will not produce sane results. 

1395 

1396 ${ARRAY_FUNCTION_LIKE} 

1397 

1398 .. versionadded:: 1.20.0 

1399 

1400 Returns 

1401 ------- 

1402 arr : ndarray 

1403 The constructed array. 

1404 

1405 Raises 

1406 ------ 

1407 ValueError 

1408 If the string is not the correct size to satisfy the requested 

1409 `dtype` and `count`. 

1410 

1411 See Also 

1412 -------- 

1413 frombuffer, fromfile, fromiter 

1414 

1415 Examples 

1416 -------- 

1417 >>> import numpy as np 

1418 >>> np.fromstring('1 2', dtype=int, sep=' ') 

1419 array([1, 2]) 

1420 >>> np.fromstring('1, 2', dtype=int, sep=',') 

1421 array([1, 2]) 

1422 

1423 """) 

1424 

1425add_newdoc('numpy._core.multiarray', 'compare_chararrays', 

1426 """ 

1427 compare_chararrays(a1, a2, cmp, rstrip) 

1428 

1429 Performs element-wise comparison of two string arrays using the 

1430 comparison operator specified by `cmp`. 

1431 

1432 Parameters 

1433 ---------- 

1434 a1, a2 : array_like 

1435 Arrays to be compared. 

1436 cmp : {"<", "<=", "==", ">=", ">", "!="} 

1437 Type of comparison. 

1438 rstrip : Boolean 

1439 If True, the spaces at the end of Strings are removed before the comparison. 

1440 

1441 Returns 

1442 ------- 

1443 out : ndarray 

1444 The output array of type Boolean with the same shape as a and b. 

1445 

1446 Raises 

1447 ------ 

1448 ValueError 

1449 If `cmp` is not valid. 

1450 TypeError 

1451 If at least one of `a` or `b` is a non-string array 

1452 

1453 Examples 

1454 -------- 

1455 >>> import numpy as np 

1456 >>> a = np.array(["a", "b", "cde"]) 

1457 >>> b = np.array(["a", "a", "dec"]) 

1458 >>> np.char.compare_chararrays(a, b, ">", True) 

1459 array([False, True, False]) 

1460 

1461 """) 

1462 

1463add_newdoc('numpy._core.multiarray', 'fromiter', 

1464 """ 

1465 fromiter(iter, dtype, count=-1, *, like=None) 

1466 

1467 Create a new 1-dimensional array from an iterable object. 

1468 

1469 Parameters 

1470 ---------- 

1471 iter : iterable object 

1472 An iterable object providing data for the array. 

1473 dtype : data-type 

1474 The data-type of the returned array. 

1475 

1476 .. versionchanged:: 1.23 

1477 Object and subarray dtypes are now supported (note that the final 

1478 result is not 1-D for a subarray dtype). 

1479 

1480 count : int, optional 

1481 The number of items to read from *iterable*. The default is -1, 

1482 which means all data is read. 

1483 ${ARRAY_FUNCTION_LIKE} 

1484 

1485 .. versionadded:: 1.20.0 

1486 

1487 Returns 

1488 ------- 

1489 out : ndarray 

1490 The output array. 

1491 

1492 Notes 

1493 ----- 

1494 Specify `count` to improve performance. It allows ``fromiter`` to 

1495 pre-allocate the output array, instead of resizing it on demand. 

1496 

1497 Examples 

1498 -------- 

1499 >>> import numpy as np 

1500 >>> iterable = (x*x for x in range(5)) 

1501 >>> np.fromiter(iterable, float) 

1502 array([ 0., 1., 4., 9., 16.]) 

1503 

1504 A carefully constructed subarray dtype will lead to higher dimensional 

1505 results: 

1506 

1507 >>> iterable = ((x+1, x+2) for x in range(5)) 

1508 >>> np.fromiter(iterable, dtype=np.dtype((int, 2))) 

1509 array([[1, 2], 

1510 [2, 3], 

1511 [3, 4], 

1512 [4, 5], 

1513 [5, 6]]) 

1514 

1515 

1516 """) 

1517 

1518add_newdoc('numpy._core.multiarray', 'fromfile', 

1519 """ 

1520 fromfile(file, dtype=float, count=-1, sep='', offset=0, *, like=None) 

1521 

1522 Construct an array from data in a text or binary file. 

1523 

1524 A highly efficient way of reading binary data with a known data-type, 

1525 as well as parsing simply formatted text files. Data written using the 

1526 `tofile` method can be read using this function. 

1527 

1528 Parameters 

1529 ---------- 

1530 file : file or str or Path 

1531 Open file object or filename. 

1532 dtype : data-type 

1533 Data type of the returned array. 

1534 For binary files, it is used to determine the size and byte-order 

1535 of the items in the file. 

1536 Most builtin numeric types are supported and extension types may be supported. 

1537 count : int 

1538 Number of items to read. ``-1`` means all items (i.e., the complete 

1539 file). 

1540 sep : str 

1541 Separator between items if file is a text file. 

1542 Empty ("") separator means the file should be treated as binary. 

1543 Spaces (" ") in the separator match zero or more whitespace characters. 

1544 A separator consisting only of spaces must match at least one 

1545 whitespace. 

1546 offset : int 

1547 The offset (in bytes) from the file's current position. Defaults to 0. 

1548 Only permitted for binary files. 

1549 ${ARRAY_FUNCTION_LIKE} 

1550 

1551 .. versionadded:: 1.20.0 

1552 

1553 See also 

1554 -------- 

1555 load, save 

1556 ndarray.tofile 

1557 loadtxt : More flexible way of loading data from a text file. 

1558 

1559 Notes 

1560 ----- 

1561 Do not rely on the combination of `tofile` and `fromfile` for 

1562 data storage, as the binary files generated are not platform 

1563 independent. In particular, no byte-order or data-type information is 

1564 saved. Data can be stored in the platform independent ``.npy`` format 

1565 using `save` and `load` instead. 

1566 

1567 Examples 

1568 -------- 

1569 Construct an ndarray: 

1570 

1571 >>> import numpy as np 

1572 >>> dt = np.dtype([('time', [('min', np.int64), ('sec', np.int64)]), 

1573 ... ('temp', float)]) 

1574 >>> x = np.zeros((1,), dtype=dt) 

1575 >>> x['time']['min'] = 10; x['temp'] = 98.25 

1576 >>> x 

1577 array([((10, 0), 98.25)], 

1578 dtype=[('time', [('min', '<i8'), ('sec', '<i8')]), ('temp', '<f8')]) 

1579 

1580 Save the raw data to disk: 

1581 

1582 >>> import tempfile 

1583 >>> fname = tempfile.mkstemp()[1] 

1584 >>> x.tofile(fname) 

1585 

1586 Read the raw data from disk: 

1587 

1588 >>> np.fromfile(fname, dtype=dt) 

1589 array([((10, 0), 98.25)], 

1590 dtype=[('time', [('min', '<i8'), ('sec', '<i8')]), ('temp', '<f8')]) 

1591 

1592 The recommended way to store and load data: 

1593 

1594 >>> np.save(fname, x) 

1595 >>> np.load(fname + '.npy') 

1596 array([((10, 0), 98.25)], 

1597 dtype=[('time', [('min', '<i8'), ('sec', '<i8')]), ('temp', '<f8')]) 

1598 

1599 """) 

1600 

1601add_newdoc('numpy._core.multiarray', 'frombuffer', 

1602 """ 

1603 frombuffer(buffer, dtype=float, count=-1, offset=0, *, like=None) 

1604 

1605 Interpret a buffer as a 1-dimensional array. 

1606 

1607 Parameters 

1608 ---------- 

1609 buffer : buffer_like 

1610 An object that exposes the buffer interface. 

1611 dtype : data-type, optional 

1612 Data-type of the returned array; default: float. 

1613 count : int, optional 

1614 Number of items to read. ``-1`` means all data in the buffer. 

1615 offset : int, optional 

1616 Start reading the buffer from this offset (in bytes); default: 0. 

1617 ${ARRAY_FUNCTION_LIKE} 

1618 

1619 .. versionadded:: 1.20.0 

1620 

1621 Returns 

1622 ------- 

1623 out : ndarray 

1624 

1625 See also 

1626 -------- 

1627 ndarray.tobytes 

1628 Inverse of this operation, construct Python bytes from the raw data 

1629 bytes in the array. 

1630 

1631 Notes 

1632 ----- 

1633 If the buffer has data that is not in machine byte-order, this should 

1634 be specified as part of the data-type, e.g.:: 

1635 

1636 >>> dt = np.dtype(int) 

1637 >>> dt = dt.newbyteorder('>') 

1638 >>> np.frombuffer(buf, dtype=dt) # doctest: +SKIP 

1639 

1640 The data of the resulting array will not be byteswapped, but will be 

1641 interpreted correctly. 

1642 

1643 This function creates a view into the original object. This should be safe 

1644 in general, but it may make sense to copy the result when the original 

1645 object is mutable or untrusted. 

1646 

1647 Examples 

1648 -------- 

1649 >>> import numpy as np 

1650 >>> s = b'hello world' 

1651 >>> np.frombuffer(s, dtype='S1', count=5, offset=6) 

1652 array([b'w', b'o', b'r', b'l', b'd'], dtype='|S1') 

1653 

1654 >>> np.frombuffer(b'\\x01\\x02', dtype=np.uint8) 

1655 array([1, 2], dtype=uint8) 

1656 >>> np.frombuffer(b'\\x01\\x02\\x03\\x04\\x05', dtype=np.uint8, count=3) 

1657 array([1, 2, 3], dtype=uint8) 

1658 

1659 """) 

1660 

1661add_newdoc('numpy._core.multiarray', 'from_dlpack', 

1662 """ 

1663 from_dlpack(x, /, *, device=None, copy=None) 

1664 

1665 Create a NumPy array from an object implementing the ``__dlpack__`` 

1666 protocol. Generally, the returned NumPy array is a read-only view 

1667 of the input object. See [1]_ and [2]_ for more details. 

1668 

1669 Parameters 

1670 ---------- 

1671 x : object 

1672 A Python object that implements the ``__dlpack__`` and 

1673 ``__dlpack_device__`` methods. 

1674 device : device, optional 

1675 Device on which to place the created array. Default: ``None``. 

1676 Must be ``"cpu"`` if passed which may allow importing an array 

1677 that is not already CPU available. 

1678 copy : bool, optional 

1679 Boolean indicating whether or not to copy the input. If ``True``, 

1680 the copy will be made. If ``False``, the function will never copy, 

1681 and will raise ``BufferError`` in case a copy is deemed necessary. 

1682 Passing it requests a copy from the exporter who may or may not 

1683 implement the capability. 

1684 If ``None``, the function will reuse the existing memory buffer if 

1685 possible and copy otherwise. Default: ``None``. 

1686 

1687 

1688 Returns 

1689 ------- 

1690 out : ndarray 

1691 

1692 References 

1693 ---------- 

1694 .. [1] Array API documentation, 

1695 https://data-apis.org/array-api/latest/design_topics/data_interchange.html#syntax-for-data-interchange-with-dlpack 

1696 

1697 .. [2] Python specification for DLPack, 

1698 https://dmlc.github.io/dlpack/latest/python_spec.html 

1699 

1700 Examples 

1701 -------- 

1702 >>> import torch # doctest: +SKIP 

1703 >>> x = torch.arange(10) # doctest: +SKIP 

1704 >>> # create a view of the torch tensor "x" in NumPy 

1705 >>> y = np.from_dlpack(x) # doctest: +SKIP 

1706 """) 

1707 

1708add_newdoc('numpy._core.multiarray', 'correlate', 

1709 """cross_correlate(a,v, mode=0)""") 

1710 

1711add_newdoc('numpy._core.multiarray', 'arange', 

1712 """ 

1713 arange([start,] stop[, step,], dtype=None, *, device=None, like=None) 

1714 

1715 Return evenly spaced values within a given interval. 

1716 

1717 ``arange`` can be called with a varying number of positional arguments: 

1718 

1719 * ``arange(stop)``: Values are generated within the half-open interval 

1720 ``[0, stop)`` (in other words, the interval including `start` but 

1721 excluding `stop`). 

1722 * ``arange(start, stop)``: Values are generated within the half-open 

1723 interval ``[start, stop)``. 

1724 * ``arange(start, stop, step)`` Values are generated within the half-open 

1725 interval ``[start, stop)``, with spacing between values given by 

1726 ``step``. 

1727 

1728 For integer arguments the function is roughly equivalent to the Python 

1729 built-in :py:class:`range`, but returns an ndarray rather than a ``range`` 

1730 instance. 

1731 

1732 When using a non-integer step, such as 0.1, it is often better to use 

1733 `numpy.linspace`. 

1734 

1735 See the Warning sections below for more information. 

1736 

1737 Parameters 

1738 ---------- 

1739 start : integer or real, optional 

1740 Start of interval. The interval includes this value. The default 

1741 start value is 0. 

1742 stop : integer or real 

1743 End of interval. The interval does not include this value, except 

1744 in some cases where `step` is not an integer and floating point 

1745 round-off affects the length of `out`. 

1746 step : integer or real, optional 

1747 Spacing between values. For any output `out`, this is the distance 

1748 between two adjacent values, ``out[i+1] - out[i]``. The default 

1749 step size is 1. If `step` is specified as a position argument, 

1750 `start` must also be given. 

1751 dtype : dtype, optional 

1752 The type of the output array. If `dtype` is not given, infer the data 

1753 type from the other input arguments. 

1754 device : str, optional 

1755 The device on which to place the created array. Default: ``None``. 

1756 For Array-API interoperability only, so must be ``"cpu"`` if passed. 

1757 

1758 .. versionadded:: 2.0.0 

1759 ${ARRAY_FUNCTION_LIKE} 

1760 

1761 .. versionadded:: 1.20.0 

1762 

1763 Returns 

1764 ------- 

1765 arange : ndarray 

1766 Array of evenly spaced values. 

1767 

1768 For floating point arguments, the length of the result is 

1769 ``ceil((stop - start)/step)``. Because of floating point overflow, 

1770 this rule may result in the last element of `out` being greater 

1771 than `stop`. 

1772 

1773 Warnings 

1774 -------- 

1775 The length of the output might not be numerically stable. 

1776 

1777 Another stability issue is due to the internal implementation of 

1778 `numpy.arange`. 

1779 The actual step value used to populate the array is 

1780 ``dtype(start + step) - dtype(start)`` and not `step`. Precision loss 

1781 can occur here, due to casting or due to using floating points when 

1782 `start` is much larger than `step`. This can lead to unexpected 

1783 behaviour. For example:: 

1784 

1785 >>> np.arange(0, 5, 0.5, dtype=int) 

1786 array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0]) 

1787 >>> np.arange(-3, 3, 0.5, dtype=int) 

1788 array([-3, -2, -1, 0, 1, 2, 3, 4, 5, 6, 7, 8]) 

1789 

1790 In such cases, the use of `numpy.linspace` should be preferred. 

1791 

1792 The built-in :py:class:`range` generates :std:doc:`Python built-in integers 

1793 that have arbitrary size <python:c-api/long>`, while `numpy.arange` 

1794 produces `numpy.int32` or `numpy.int64` numbers. This may result in 

1795 incorrect results for large integer values:: 

1796 

1797 >>> power = 40 

1798 >>> modulo = 10000 

1799 >>> x1 = [(n ** power) % modulo for n in range(8)] 

1800 >>> x2 = [(n ** power) % modulo for n in np.arange(8)] 

1801 >>> print(x1) 

1802 [0, 1, 7776, 8801, 6176, 625, 6576, 4001] # correct 

1803 >>> print(x2) 

1804 [0, 1, 7776, 7185, 0, 5969, 4816, 3361] # incorrect 

1805 

1806 See Also 

1807 -------- 

1808 numpy.linspace : Evenly spaced numbers with careful handling of endpoints. 

1809 numpy.ogrid: Arrays of evenly spaced numbers in N-dimensions. 

1810 numpy.mgrid: Grid-shaped arrays of evenly spaced numbers in N-dimensions. 

1811 :ref:`how-to-partition` 

1812 

1813 Examples 

1814 -------- 

1815 >>> import numpy as np 

1816 >>> np.arange(3) 

1817 array([0, 1, 2]) 

1818 >>> np.arange(3.0) 

1819 array([ 0., 1., 2.]) 

1820 >>> np.arange(3,7) 

1821 array([3, 4, 5, 6]) 

1822 >>> np.arange(3,7,2) 

1823 array([3, 5]) 

1824 

1825 """) 

1826 

1827add_newdoc('numpy._core.multiarray', '_get_ndarray_c_version', 

1828 """_get_ndarray_c_version() 

1829 

1830 Return the compile time NPY_VERSION (formerly called NDARRAY_VERSION) number. 

1831 

1832 """) 

1833 

1834add_newdoc('numpy._core.multiarray', '_reconstruct', 

1835 """_reconstruct(subtype, shape, dtype) 

1836 

1837 Construct an empty array. Used by Pickles. 

1838 

1839 """) 

1840 

1841add_newdoc('numpy._core.multiarray', 'promote_types', 

1842 """ 

1843 promote_types(type1, type2) 

1844 

1845 Returns the data type with the smallest size and smallest scalar 

1846 kind to which both ``type1`` and ``type2`` may be safely cast. 

1847 The returned data type is always considered "canonical", this mainly 

1848 means that the promoted dtype will always be in native byte order. 

1849 

1850 This function is symmetric, but rarely associative. 

1851 

1852 Parameters 

1853 ---------- 

1854 type1 : dtype or dtype specifier 

1855 First data type. 

1856 type2 : dtype or dtype specifier 

1857 Second data type. 

1858 

1859 Returns 

1860 ------- 

1861 out : dtype 

1862 The promoted data type. 

1863 

1864 Notes 

1865 ----- 

1866 Please see `numpy.result_type` for additional information about promotion. 

1867 

1868 Starting in NumPy 1.9, promote_types function now returns a valid string 

1869 length when given an integer or float dtype as one argument and a string 

1870 dtype as another argument. Previously it always returned the input string 

1871 dtype, even if it wasn't long enough to store the max integer/float value 

1872 converted to a string. 

1873 

1874 .. versionchanged:: 1.23.0 

1875 

1876 NumPy now supports promotion for more structured dtypes. It will now 

1877 remove unnecessary padding from a structure dtype and promote included 

1878 fields individually. 

1879 

1880 See Also 

1881 -------- 

1882 result_type, dtype, can_cast 

1883 

1884 Examples 

1885 -------- 

1886 >>> import numpy as np 

1887 >>> np.promote_types('f4', 'f8') 

1888 dtype('float64') 

1889 

1890 >>> np.promote_types('i8', 'f4') 

1891 dtype('float64') 

1892 

1893 >>> np.promote_types('>i8', '<c8') 

1894 dtype('complex128') 

1895 

1896 >>> np.promote_types('i4', 'S8') 

1897 dtype('S11') 

1898 

1899 An example of a non-associative case: 

1900 

1901 >>> p = np.promote_types 

1902 >>> p('S', p('i1', 'u1')) 

1903 dtype('S6') 

1904 >>> p(p('S', 'i1'), 'u1') 

1905 dtype('S4') 

1906 

1907 """) 

1908 

1909add_newdoc('numpy._core.multiarray', 'c_einsum', 

1910 """ 

1911 c_einsum(subscripts, *operands, out=None, dtype=None, order='K', 

1912 casting='safe') 

1913 

1914 *This documentation shadows that of the native python implementation of the `einsum` function, 

1915 except all references and examples related to the `optimize` argument (v 0.12.0) have been removed.* 

1916 

1917 Evaluates the Einstein summation convention on the operands. 

1918 

1919 Using the Einstein summation convention, many common multi-dimensional, 

1920 linear algebraic array operations can be represented in a simple fashion. 

1921 In *implicit* mode `einsum` computes these values. 

1922 

1923 In *explicit* mode, `einsum` provides further flexibility to compute 

1924 other array operations that might not be considered classical Einstein 

1925 summation operations, by disabling, or forcing summation over specified 

1926 subscript labels. 

1927 

1928 See the notes and examples for clarification. 

1929 

1930 Parameters 

1931 ---------- 

1932 subscripts : str 

1933 Specifies the subscripts for summation as comma separated list of 

1934 subscript labels. An implicit (classical Einstein summation) 

1935 calculation is performed unless the explicit indicator '->' is 

1936 included as well as subscript labels of the precise output form. 

1937 operands : list of array_like 

1938 These are the arrays for the operation. 

1939 out : ndarray, optional 

1940 If provided, the calculation is done into this array. 

1941 dtype : {data-type, None}, optional 

1942 If provided, forces the calculation to use the data type specified. 

1943 Note that you may have to also give a more liberal `casting` 

1944 parameter to allow the conversions. Default is None. 

1945 order : {'C', 'F', 'A', 'K'}, optional 

1946 Controls the memory layout of the output. 'C' means it should 

1947 be C contiguous. 'F' means it should be Fortran contiguous, 

1948 'A' means it should be 'F' if the inputs are all 'F', 'C' otherwise. 

1949 'K' means it should be as close to the layout of the inputs as 

1950 is possible, including arbitrarily permuted axes. 

1951 Default is 'K'. 

1952 casting : {'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional 

1953 Controls what kind of data casting may occur. Setting this to 

1954 'unsafe' is not recommended, as it can adversely affect accumulations. 

1955 

1956 * 'no' means the data types should not be cast at all. 

1957 * 'equiv' means only byte-order changes are allowed. 

1958 * 'safe' means only casts which can preserve values are allowed. 

1959 * 'same_kind' means only safe casts or casts within a kind, 

1960 like float64 to float32, are allowed. 

1961 * 'unsafe' means any data conversions may be done. 

1962 

1963 Default is 'safe'. 

1964 optimize : {False, True, 'greedy', 'optimal'}, optional 

1965 Controls if intermediate optimization should occur. No optimization 

1966 will occur if False and True will default to the 'greedy' algorithm. 

1967 Also accepts an explicit contraction list from the ``np.einsum_path`` 

1968 function. See ``np.einsum_path`` for more details. Defaults to False. 

1969 

1970 Returns 

1971 ------- 

1972 output : ndarray 

1973 The calculation based on the Einstein summation convention. 

1974 

1975 See Also 

1976 -------- 

1977 einsum_path, dot, inner, outer, tensordot, linalg.multi_dot 

1978 

1979 Notes 

1980 ----- 

1981 The Einstein summation convention can be used to compute 

1982 many multi-dimensional, linear algebraic array operations. `einsum` 

1983 provides a succinct way of representing these. 

1984 

1985 A non-exhaustive list of these operations, 

1986 which can be computed by `einsum`, is shown below along with examples: 

1987 

1988 * Trace of an array, :py:func:`numpy.trace`. 

1989 * Return a diagonal, :py:func:`numpy.diag`. 

1990 * Array axis summations, :py:func:`numpy.sum`. 

1991 * Transpositions and permutations, :py:func:`numpy.transpose`. 

1992 * Matrix multiplication and dot product, :py:func:`numpy.matmul` :py:func:`numpy.dot`. 

1993 * Vector inner and outer products, :py:func:`numpy.inner` :py:func:`numpy.outer`. 

1994 * Broadcasting, element-wise and scalar multiplication, :py:func:`numpy.multiply`. 

1995 * Tensor contractions, :py:func:`numpy.tensordot`. 

1996 * Chained array operations, in efficient calculation order, :py:func:`numpy.einsum_path`. 

1997 

1998 The subscripts string is a comma-separated list of subscript labels, 

1999 where each label refers to a dimension of the corresponding operand. 

2000 Whenever a label is repeated it is summed, so ``np.einsum('i,i', a, b)`` 

2001 is equivalent to :py:func:`np.inner(a,b) <numpy.inner>`. If a label 

2002 appears only once, it is not summed, so ``np.einsum('i', a)`` produces a 

2003 view of ``a`` with no changes. A further example ``np.einsum('ij,jk', a, b)`` 

2004 describes traditional matrix multiplication and is equivalent to 

2005 :py:func:`np.matmul(a,b) <numpy.matmul>`. Repeated subscript labels in one 

2006 operand take the diagonal. For example, ``np.einsum('ii', a)`` is equivalent 

2007 to :py:func:`np.trace(a) <numpy.trace>`. 

2008 

2009 In *implicit mode*, the chosen subscripts are important 

2010 since the axes of the output are reordered alphabetically. This 

2011 means that ``np.einsum('ij', a)`` doesn't affect a 2D array, while 

2012 ``np.einsum('ji', a)`` takes its transpose. Additionally, 

2013 ``np.einsum('ij,jk', a, b)`` returns a matrix multiplication, while, 

2014 ``np.einsum('ij,jh', a, b)`` returns the transpose of the 

2015 multiplication since subscript 'h' precedes subscript 'i'. 

2016 

2017 In *explicit mode* the output can be directly controlled by 

2018 specifying output subscript labels. This requires the 

2019 identifier '->' as well as the list of output subscript labels. 

2020 This feature increases the flexibility of the function since 

2021 summing can be disabled or forced when required. The call 

2022 ``np.einsum('i->', a)`` is like :py:func:`np.sum(a) <numpy.sum>` 

2023 if ``a`` is a 1-D array, and ``np.einsum('ii->i', a)`` 

2024 is like :py:func:`np.diag(a) <numpy.diag>` if ``a`` is a square 2-D array. 

2025 The difference is that `einsum` does not allow broadcasting by default. 

2026 Additionally ``np.einsum('ij,jh->ih', a, b)`` directly specifies the 

2027 order of the output subscript labels and therefore returns matrix 

2028 multiplication, unlike the example above in implicit mode. 

2029 

2030 To enable and control broadcasting, use an ellipsis. Default 

2031 NumPy-style broadcasting is done by adding an ellipsis 

2032 to the left of each term, like ``np.einsum('...ii->...i', a)``. 

2033 ``np.einsum('...i->...', a)`` is like 

2034 :py:func:`np.sum(a, axis=-1) <numpy.sum>` for array ``a`` of any shape. 

2035 To take the trace along the first and last axes, 

2036 you can do ``np.einsum('i...i', a)``, or to do a matrix-matrix 

2037 product with the left-most indices instead of rightmost, one can do 

2038 ``np.einsum('ij...,jk...->ik...', a, b)``. 

2039 

2040 When there is only one operand, no axes are summed, and no output 

2041 parameter is provided, a view into the operand is returned instead 

2042 of a new array. Thus, taking the diagonal as ``np.einsum('ii->i', a)`` 

2043 produces a view (changed in version 1.10.0). 

2044 

2045 `einsum` also provides an alternative way to provide the subscripts 

2046 and operands as ``einsum(op0, sublist0, op1, sublist1, ..., [sublistout])``. 

2047 If the output shape is not provided in this format `einsum` will be 

2048 calculated in implicit mode, otherwise it will be performed explicitly. 

2049 The examples below have corresponding `einsum` calls with the two 

2050 parameter methods. 

2051 

2052 Views returned from einsum are now writeable whenever the input array 

2053 is writeable. For example, ``np.einsum('ijk...->kji...', a)`` will now 

2054 have the same effect as :py:func:`np.swapaxes(a, 0, 2) <numpy.swapaxes>` 

2055 and ``np.einsum('ii->i', a)`` will return a writeable view of the diagonal 

2056 of a 2D array. 

2057 

2058 Examples 

2059 -------- 

2060 >>> import numpy as np 

2061 >>> a = np.arange(25).reshape(5,5) 

2062 >>> b = np.arange(5) 

2063 >>> c = np.arange(6).reshape(2,3) 

2064 

2065 Trace of a matrix: 

2066 

2067 >>> np.einsum('ii', a) 

2068 60 

2069 >>> np.einsum(a, [0,0]) 

2070 60 

2071 >>> np.trace(a) 

2072 60 

2073 

2074 Extract the diagonal (requires explicit form): 

2075 

2076 >>> np.einsum('ii->i', a) 

2077 array([ 0, 6, 12, 18, 24]) 

2078 >>> np.einsum(a, [0,0], [0]) 

2079 array([ 0, 6, 12, 18, 24]) 

2080 >>> np.diag(a) 

2081 array([ 0, 6, 12, 18, 24]) 

2082 

2083 Sum over an axis (requires explicit form): 

2084 

2085 >>> np.einsum('ij->i', a) 

2086 array([ 10, 35, 60, 85, 110]) 

2087 >>> np.einsum(a, [0,1], [0]) 

2088 array([ 10, 35, 60, 85, 110]) 

2089 >>> np.sum(a, axis=1) 

2090 array([ 10, 35, 60, 85, 110]) 

2091 

2092 For higher dimensional arrays summing a single axis can be done with ellipsis: 

2093 

2094 >>> np.einsum('...j->...', a) 

2095 array([ 10, 35, 60, 85, 110]) 

2096 >>> np.einsum(a, [Ellipsis,1], [Ellipsis]) 

2097 array([ 10, 35, 60, 85, 110]) 

2098 

2099 Compute a matrix transpose, or reorder any number of axes: 

2100 

2101 >>> np.einsum('ji', c) 

2102 array([[0, 3], 

2103 [1, 4], 

2104 [2, 5]]) 

2105 >>> np.einsum('ij->ji', c) 

2106 array([[0, 3], 

2107 [1, 4], 

2108 [2, 5]]) 

2109 >>> np.einsum(c, [1,0]) 

2110 array([[0, 3], 

2111 [1, 4], 

2112 [2, 5]]) 

2113 >>> np.transpose(c) 

2114 array([[0, 3], 

2115 [1, 4], 

2116 [2, 5]]) 

2117 

2118 Vector inner products: 

2119 

2120 >>> np.einsum('i,i', b, b) 

2121 30 

2122 >>> np.einsum(b, [0], b, [0]) 

2123 30 

2124 >>> np.inner(b,b) 

2125 30 

2126 

2127 Matrix vector multiplication: 

2128 

2129 >>> np.einsum('ij,j', a, b) 

2130 array([ 30, 80, 130, 180, 230]) 

2131 >>> np.einsum(a, [0,1], b, [1]) 

2132 array([ 30, 80, 130, 180, 230]) 

2133 >>> np.dot(a, b) 

2134 array([ 30, 80, 130, 180, 230]) 

2135 >>> np.einsum('...j,j', a, b) 

2136 array([ 30, 80, 130, 180, 230]) 

2137 

2138 Broadcasting and scalar multiplication: 

2139 

2140 >>> np.einsum('..., ...', 3, c) 

2141 array([[ 0, 3, 6], 

2142 [ 9, 12, 15]]) 

2143 >>> np.einsum(',ij', 3, c) 

2144 array([[ 0, 3, 6], 

2145 [ 9, 12, 15]]) 

2146 >>> np.einsum(3, [Ellipsis], c, [Ellipsis]) 

2147 array([[ 0, 3, 6], 

2148 [ 9, 12, 15]]) 

2149 >>> np.multiply(3, c) 

2150 array([[ 0, 3, 6], 

2151 [ 9, 12, 15]]) 

2152 

2153 Vector outer product: 

2154 

2155 >>> np.einsum('i,j', np.arange(2)+1, b) 

2156 array([[0, 1, 2, 3, 4], 

2157 [0, 2, 4, 6, 8]]) 

2158 >>> np.einsum(np.arange(2)+1, [0], b, [1]) 

2159 array([[0, 1, 2, 3, 4], 

2160 [0, 2, 4, 6, 8]]) 

2161 >>> np.outer(np.arange(2)+1, b) 

2162 array([[0, 1, 2, 3, 4], 

2163 [0, 2, 4, 6, 8]]) 

2164 

2165 Tensor contraction: 

2166 

2167 >>> a = np.arange(60.).reshape(3,4,5) 

2168 >>> b = np.arange(24.).reshape(4,3,2) 

2169 >>> np.einsum('ijk,jil->kl', a, b) 

2170 array([[ 4400., 4730.], 

2171 [ 4532., 4874.], 

2172 [ 4664., 5018.], 

2173 [ 4796., 5162.], 

2174 [ 4928., 5306.]]) 

2175 >>> np.einsum(a, [0,1,2], b, [1,0,3], [2,3]) 

2176 array([[ 4400., 4730.], 

2177 [ 4532., 4874.], 

2178 [ 4664., 5018.], 

2179 [ 4796., 5162.], 

2180 [ 4928., 5306.]]) 

2181 >>> np.tensordot(a,b, axes=([1,0],[0,1])) 

2182 array([[ 4400., 4730.], 

2183 [ 4532., 4874.], 

2184 [ 4664., 5018.], 

2185 [ 4796., 5162.], 

2186 [ 4928., 5306.]]) 

2187 

2188 Writeable returned arrays (since version 1.10.0): 

2189 

2190 >>> a = np.zeros((3, 3)) 

2191 >>> np.einsum('ii->i', a)[:] = 1 

2192 >>> a 

2193 array([[ 1., 0., 0.], 

2194 [ 0., 1., 0.], 

2195 [ 0., 0., 1.]]) 

2196 

2197 Example of ellipsis use: 

2198 

2199 >>> a = np.arange(6).reshape((3,2)) 

2200 >>> b = np.arange(12).reshape((4,3)) 

2201 >>> np.einsum('ki,jk->ij', a, b) 

2202 array([[10, 28, 46, 64], 

2203 [13, 40, 67, 94]]) 

2204 >>> np.einsum('ki,...k->i...', a, b) 

2205 array([[10, 28, 46, 64], 

2206 [13, 40, 67, 94]]) 

2207 >>> np.einsum('k...,jk', a, b) 

2208 array([[10, 28, 46, 64], 

2209 [13, 40, 67, 94]]) 

2210 

2211 """) 

2212 

2213 

2214############################################################################## 

2215# 

2216# Documentation for ndarray attributes and methods 

2217# 

2218############################################################################## 

2219 

2220 

2221############################################################################## 

2222# 

2223# ndarray object 

2224# 

2225############################################################################## 

2226 

2227 

2228add_newdoc('numpy._core.multiarray', 'ndarray', 

2229 """ 

2230 ndarray(shape, dtype=float, buffer=None, offset=0, 

2231 strides=None, order=None) 

2232 

2233 An array object represents a multidimensional, homogeneous array 

2234 of fixed-size items. An associated data-type object describes the 

2235 format of each element in the array (its byte-order, how many bytes it 

2236 occupies in memory, whether it is an integer, a floating point number, 

2237 or something else, etc.) 

2238 

2239 Arrays should be constructed using `array`, `zeros` or `empty` (refer 

2240 to the See Also section below). The parameters given here refer to 

2241 a low-level method (`ndarray(...)`) for instantiating an array. 

2242 

2243 For more information, refer to the `numpy` module and examine the 

2244 methods and attributes of an array. 

2245 

2246 Parameters 

2247 ---------- 

2248 (for the __new__ method; see Notes below) 

2249 

2250 shape : tuple of ints 

2251 Shape of created array. 

2252 dtype : data-type, optional 

2253 Any object that can be interpreted as a numpy data type. 

2254 buffer : object exposing buffer interface, optional 

2255 Used to fill the array with data. 

2256 offset : int, optional 

2257 Offset of array data in buffer. 

2258 strides : tuple of ints, optional 

2259 Strides of data in memory. 

2260 order : {'C', 'F'}, optional 

2261 Row-major (C-style) or column-major (Fortran-style) order. 

2262 

2263 Attributes 

2264 ---------- 

2265 T : ndarray 

2266 Transpose of the array. 

2267 data : buffer 

2268 The array's elements, in memory. 

2269 dtype : dtype object 

2270 Describes the format of the elements in the array. 

2271 flags : dict 

2272 Dictionary containing information related to memory use, e.g., 

2273 'C_CONTIGUOUS', 'OWNDATA', 'WRITEABLE', etc. 

2274 flat : numpy.flatiter object 

2275 Flattened version of the array as an iterator. The iterator 

2276 allows assignments, e.g., ``x.flat = 3`` (See `ndarray.flat` for 

2277 assignment examples; TODO). 

2278 imag : ndarray 

2279 Imaginary part of the array. 

2280 real : ndarray 

2281 Real part of the array. 

2282 size : int 

2283 Number of elements in the array. 

2284 itemsize : int 

2285 The memory use of each array element in bytes. 

2286 nbytes : int 

2287 The total number of bytes required to store the array data, 

2288 i.e., ``itemsize * size``. 

2289 ndim : int 

2290 The array's number of dimensions. 

2291 shape : tuple of ints 

2292 Shape of the array. 

2293 strides : tuple of ints 

2294 The step-size required to move from one element to the next in 

2295 memory. For example, a contiguous ``(3, 4)`` array of type 

2296 ``int16`` in C-order has strides ``(8, 2)``. This implies that 

2297 to move from element to element in memory requires jumps of 2 bytes. 

2298 To move from row-to-row, one needs to jump 8 bytes at a time 

2299 (``2 * 4``). 

2300 ctypes : ctypes object 

2301 Class containing properties of the array needed for interaction 

2302 with ctypes. 

2303 base : ndarray 

2304 If the array is a view into another array, that array is its `base` 

2305 (unless that array is also a view). The `base` array is where the 

2306 array data is actually stored. 

2307 

2308 See Also 

2309 -------- 

2310 array : Construct an array. 

2311 zeros : Create an array, each element of which is zero. 

2312 empty : Create an array, but leave its allocated memory unchanged (i.e., 

2313 it contains "garbage"). 

2314 dtype : Create a data-type. 

2315 numpy.typing.NDArray : An ndarray alias :term:`generic <generic type>` 

2316 w.r.t. its `dtype.type <numpy.dtype.type>`. 

2317 

2318 Notes 

2319 ----- 

2320 There are two modes of creating an array using ``__new__``: 

2321 

2322 1. If `buffer` is None, then only `shape`, `dtype`, and `order` 

2323 are used. 

2324 2. If `buffer` is an object exposing the buffer interface, then 

2325 all keywords are interpreted. 

2326 

2327 No ``__init__`` method is needed because the array is fully initialized 

2328 after the ``__new__`` method. 

2329 

2330 Examples 

2331 -------- 

2332 These examples illustrate the low-level `ndarray` constructor. Refer 

2333 to the `See Also` section above for easier ways of constructing an 

2334 ndarray. 

2335 

2336 First mode, `buffer` is None: 

2337 

2338 >>> import numpy as np 

2339 >>> np.ndarray(shape=(2,2), dtype=float, order='F') 

2340 array([[0.0e+000, 0.0e+000], # random 

2341 [ nan, 2.5e-323]]) 

2342 

2343 Second mode: 

2344 

2345 >>> np.ndarray((2,), buffer=np.array([1,2,3]), 

2346 ... offset=np.int_().itemsize, 

2347 ... dtype=int) # offset = 1*itemsize, i.e. skip first element 

2348 array([2, 3]) 

2349 

2350 """) 

2351 

2352 

2353############################################################################## 

2354# 

2355# ndarray attributes 

2356# 

2357############################################################################## 

2358 

2359 

2360add_newdoc('numpy._core.multiarray', 'ndarray', ('__array_interface__', 

2361 """Array protocol: Python side.""")) 

2362 

2363 

2364add_newdoc('numpy._core.multiarray', 'ndarray', ('__array_priority__', 

2365 """Array priority.""")) 

2366 

2367 

2368add_newdoc('numpy._core.multiarray', 'ndarray', ('__array_struct__', 

2369 """Array protocol: C-struct side.""")) 

2370 

2371add_newdoc('numpy._core.multiarray', 'ndarray', ('__dlpack__', 

2372 """ 

2373 a.__dlpack__(*, stream=None, max_version=None, dl_device=None, copy=None) 

2374 

2375 DLPack Protocol: Part of the Array API. 

2376 

2377 """)) 

2378 

2379add_newdoc('numpy._core.multiarray', 'ndarray', ('__dlpack_device__', 

2380 """ 

2381 a.__dlpack_device__() 

2382 

2383 DLPack Protocol: Part of the Array API. 

2384 

2385 """)) 

2386 

2387add_newdoc('numpy._core.multiarray', 'ndarray', ('base', 

2388 """ 

2389 Base object if memory is from some other object. 

2390 

2391 Examples 

2392 -------- 

2393 The base of an array that owns its memory is None: 

2394 

2395 >>> import numpy as np 

2396 >>> x = np.array([1,2,3,4]) 

2397 >>> x.base is None 

2398 True 

2399 

2400 Slicing creates a view, whose memory is shared with x: 

2401 

2402 >>> y = x[2:] 

2403 >>> y.base is x 

2404 True 

2405 

2406 """)) 

2407 

2408 

2409add_newdoc('numpy._core.multiarray', 'ndarray', ('ctypes', 

2410 """ 

2411 An object to simplify the interaction of the array with the ctypes 

2412 module. 

2413 

2414 This attribute creates an object that makes it easier to use arrays 

2415 when calling shared libraries with the ctypes module. The returned 

2416 object has, among others, data, shape, and strides attributes (see 

2417 Notes below) which themselves return ctypes objects that can be used 

2418 as arguments to a shared library. 

2419 

2420 Parameters 

2421 ---------- 

2422 None 

2423 

2424 Returns 

2425 ------- 

2426 c : Python object 

2427 Possessing attributes data, shape, strides, etc. 

2428 

2429 See Also 

2430 -------- 

2431 numpy.ctypeslib 

2432 

2433 Notes 

2434 ----- 

2435 Below are the public attributes of this object which were documented 

2436 in "Guide to NumPy" (we have omitted undocumented public attributes, 

2437 as well as documented private attributes): 

2438 

2439 .. autoattribute:: numpy._core._internal._ctypes.data 

2440 :noindex: 

2441 

2442 .. autoattribute:: numpy._core._internal._ctypes.shape 

2443 :noindex: 

2444 

2445 .. autoattribute:: numpy._core._internal._ctypes.strides 

2446 :noindex: 

2447 

2448 .. automethod:: numpy._core._internal._ctypes.data_as 

2449 :noindex: 

2450 

2451 .. automethod:: numpy._core._internal._ctypes.shape_as 

2452 :noindex: 

2453 

2454 .. automethod:: numpy._core._internal._ctypes.strides_as 

2455 :noindex: 

2456 

2457 If the ctypes module is not available, then the ctypes attribute 

2458 of array objects still returns something useful, but ctypes objects 

2459 are not returned and errors may be raised instead. In particular, 

2460 the object will still have the ``as_parameter`` attribute which will 

2461 return an integer equal to the data attribute. 

2462 

2463 Examples 

2464 -------- 

2465 >>> import numpy as np 

2466 >>> import ctypes 

2467 >>> x = np.array([[0, 1], [2, 3]], dtype=np.int32) 

2468 >>> x 

2469 array([[0, 1], 

2470 [2, 3]], dtype=int32) 

2471 >>> x.ctypes.data 

2472 31962608 # may vary 

2473 >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint32)) 

2474 <__main__.LP_c_uint object at 0x7ff2fc1fc200> # may vary 

2475 >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint32)).contents 

2476 c_uint(0) 

2477 >>> x.ctypes.data_as(ctypes.POINTER(ctypes.c_uint64)).contents 

2478 c_ulong(4294967296) 

2479 >>> x.ctypes.shape 

2480 <numpy._core._internal.c_long_Array_2 object at 0x7ff2fc1fce60> # may vary 

2481 >>> x.ctypes.strides 

2482 <numpy._core._internal.c_long_Array_2 object at 0x7ff2fc1ff320> # may vary 

2483 

2484 """)) 

2485 

2486 

2487add_newdoc('numpy._core.multiarray', 'ndarray', ('data', 

2488 """Python buffer object pointing to the start of the array's data.""")) 

2489 

2490 

2491add_newdoc('numpy._core.multiarray', 'ndarray', ('dtype', 

2492 """ 

2493 Data-type of the array's elements. 

2494 

2495 .. warning:: 

2496 

2497 Setting ``arr.dtype`` is discouraged and may be deprecated in the 

2498 future. Setting will replace the ``dtype`` without modifying the 

2499 memory (see also `ndarray.view` and `ndarray.astype`). 

2500 

2501 Parameters 

2502 ---------- 

2503 None 

2504 

2505 Returns 

2506 ------- 

2507 d : numpy dtype object 

2508 

2509 See Also 

2510 -------- 

2511 ndarray.astype : Cast the values contained in the array to a new data-type. 

2512 ndarray.view : Create a view of the same data but a different data-type. 

2513 numpy.dtype 

2514 

2515 Examples 

2516 -------- 

2517 >>> x 

2518 array([[0, 1], 

2519 [2, 3]]) 

2520 >>> x.dtype 

2521 dtype('int32') 

2522 >>> type(x.dtype) 

2523 <type 'numpy.dtype'> 

2524 

2525 """)) 

2526 

2527 

2528add_newdoc('numpy._core.multiarray', 'ndarray', ('imag', 

2529 """ 

2530 The imaginary part of the array. 

2531 

2532 Examples 

2533 -------- 

2534 >>> import numpy as np 

2535 >>> x = np.sqrt([1+0j, 0+1j]) 

2536 >>> x.imag 

2537 array([ 0. , 0.70710678]) 

2538 >>> x.imag.dtype 

2539 dtype('float64') 

2540 

2541 """)) 

2542 

2543 

2544add_newdoc('numpy._core.multiarray', 'ndarray', ('itemsize', 

2545 """ 

2546 Length of one array element in bytes. 

2547 

2548 Examples 

2549 -------- 

2550 >>> import numpy as np 

2551 >>> x = np.array([1,2,3], dtype=np.float64) 

2552 >>> x.itemsize 

2553 8 

2554 >>> x = np.array([1,2,3], dtype=np.complex128) 

2555 >>> x.itemsize 

2556 16 

2557 

2558 """)) 

2559 

2560 

2561add_newdoc('numpy._core.multiarray', 'ndarray', ('flags', 

2562 """ 

2563 Information about the memory layout of the array. 

2564 

2565 Attributes 

2566 ---------- 

2567 C_CONTIGUOUS (C) 

2568 The data is in a single, C-style contiguous segment. 

2569 F_CONTIGUOUS (F) 

2570 The data is in a single, Fortran-style contiguous segment. 

2571 OWNDATA (O) 

2572 The array owns the memory it uses or borrows it from another object. 

2573 WRITEABLE (W) 

2574 The data area can be written to. Setting this to False locks 

2575 the data, making it read-only. A view (slice, etc.) inherits WRITEABLE 

2576 from its base array at creation time, but a view of a writeable 

2577 array may be subsequently locked while the base array remains writeable. 

2578 (The opposite is not true, in that a view of a locked array may not 

2579 be made writeable. However, currently, locking a base object does not 

2580 lock any views that already reference it, so under that circumstance it 

2581 is possible to alter the contents of a locked array via a previously 

2582 created writeable view onto it.) Attempting to change a non-writeable 

2583 array raises a RuntimeError exception. 

2584 ALIGNED (A) 

2585 The data and all elements are aligned appropriately for the hardware. 

2586 WRITEBACKIFCOPY (X) 

2587 This array is a copy of some other array. The C-API function 

2588 PyArray_ResolveWritebackIfCopy must be called before deallocating 

2589 to the base array will be updated with the contents of this array. 

2590 FNC 

2591 F_CONTIGUOUS and not C_CONTIGUOUS. 

2592 FORC 

2593 F_CONTIGUOUS or C_CONTIGUOUS (one-segment test). 

2594 BEHAVED (B) 

2595 ALIGNED and WRITEABLE. 

2596 CARRAY (CA) 

2597 BEHAVED and C_CONTIGUOUS. 

2598 FARRAY (FA) 

2599 BEHAVED and F_CONTIGUOUS and not C_CONTIGUOUS. 

2600 

2601 Notes 

2602 ----- 

2603 The `flags` object can be accessed dictionary-like (as in ``a.flags['WRITEABLE']``), 

2604 or by using lowercased attribute names (as in ``a.flags.writeable``). Short flag 

2605 names are only supported in dictionary access. 

2606 

2607 Only the WRITEBACKIFCOPY, WRITEABLE, and ALIGNED flags can be 

2608 changed by the user, via direct assignment to the attribute or dictionary 

2609 entry, or by calling `ndarray.setflags`. 

2610 

2611 The array flags cannot be set arbitrarily: 

2612 

2613 - WRITEBACKIFCOPY can only be set ``False``. 

2614 - ALIGNED can only be set ``True`` if the data is truly aligned. 

2615 - WRITEABLE can only be set ``True`` if the array owns its own memory 

2616 or the ultimate owner of the memory exposes a writeable buffer 

2617 interface or is a string. 

2618 

2619 Arrays can be both C-style and Fortran-style contiguous simultaneously. 

2620 This is clear for 1-dimensional arrays, but can also be true for higher 

2621 dimensional arrays. 

2622 

2623 Even for contiguous arrays a stride for a given dimension 

2624 ``arr.strides[dim]`` may be *arbitrary* if ``arr.shape[dim] == 1`` 

2625 or the array has no elements. 

2626 It does *not* generally hold that ``self.strides[-1] == self.itemsize`` 

2627 for C-style contiguous arrays or ``self.strides[0] == self.itemsize`` for 

2628 Fortran-style contiguous arrays is true. 

2629 """)) 

2630 

2631 

2632add_newdoc('numpy._core.multiarray', 'ndarray', ('flat', 

2633 """ 

2634 A 1-D iterator over the array. 

2635 

2636 This is a `numpy.flatiter` instance, which acts similarly to, but is not 

2637 a subclass of, Python's built-in iterator object. 

2638 

2639 See Also 

2640 -------- 

2641 flatten : Return a copy of the array collapsed into one dimension. 

2642 

2643 flatiter 

2644 

2645 Examples 

2646 -------- 

2647 >>> import numpy as np 

2648 >>> x = np.arange(1, 7).reshape(2, 3) 

2649 >>> x 

2650 array([[1, 2, 3], 

2651 [4, 5, 6]]) 

2652 >>> x.flat[3] 

2653 4 

2654 >>> x.T 

2655 array([[1, 4], 

2656 [2, 5], 

2657 [3, 6]]) 

2658 >>> x.T.flat[3] 

2659 5 

2660 >>> type(x.flat) 

2661 <class 'numpy.flatiter'> 

2662 

2663 An assignment example: 

2664 

2665 >>> x.flat = 3; x 

2666 array([[3, 3, 3], 

2667 [3, 3, 3]]) 

2668 >>> x.flat[[1,4]] = 1; x 

2669 array([[3, 1, 3], 

2670 [3, 1, 3]]) 

2671 

2672 """)) 

2673 

2674 

2675add_newdoc('numpy._core.multiarray', 'ndarray', ('nbytes', 

2676 """ 

2677 Total bytes consumed by the elements of the array. 

2678 

2679 Notes 

2680 ----- 

2681 Does not include memory consumed by non-element attributes of the 

2682 array object. 

2683 

2684 See Also 

2685 -------- 

2686 sys.getsizeof 

2687 Memory consumed by the object itself without parents in case view. 

2688 This does include memory consumed by non-element attributes. 

2689 

2690 Examples 

2691 -------- 

2692 >>> import numpy as np 

2693 >>> x = np.zeros((3,5,2), dtype=np.complex128) 

2694 >>> x.nbytes 

2695 480 

2696 >>> np.prod(x.shape) * x.itemsize 

2697 480 

2698 

2699 """)) 

2700 

2701 

2702add_newdoc('numpy._core.multiarray', 'ndarray', ('ndim', 

2703 """ 

2704 Number of array dimensions. 

2705 

2706 Examples 

2707 -------- 

2708 >>> import numpy as np 

2709 >>> x = np.array([1, 2, 3]) 

2710 >>> x.ndim 

2711 1 

2712 >>> y = np.zeros((2, 3, 4)) 

2713 >>> y.ndim 

2714 3 

2715 

2716 """)) 

2717 

2718 

2719add_newdoc('numpy._core.multiarray', 'ndarray', ('real', 

2720 """ 

2721 The real part of the array. 

2722 

2723 Examples 

2724 -------- 

2725 >>> import numpy as np 

2726 >>> x = np.sqrt([1+0j, 0+1j]) 

2727 >>> x.real 

2728 array([ 1. , 0.70710678]) 

2729 >>> x.real.dtype 

2730 dtype('float64') 

2731 

2732 See Also 

2733 -------- 

2734 numpy.real : equivalent function 

2735 

2736 """)) 

2737 

2738 

2739add_newdoc('numpy._core.multiarray', 'ndarray', ('shape', 

2740 """ 

2741 Tuple of array dimensions. 

2742 

2743 The shape property is usually used to get the current shape of an array, 

2744 but may also be used to reshape the array in-place by assigning a tuple of 

2745 array dimensions to it. As with `numpy.reshape`, one of the new shape 

2746 dimensions can be -1, in which case its value is inferred from the size of 

2747 the array and the remaining dimensions. Reshaping an array in-place will 

2748 fail if a copy is required. 

2749 

2750 .. warning:: 

2751 

2752 Setting ``arr.shape`` is discouraged and may be deprecated in the 

2753 future. Using `ndarray.reshape` is the preferred approach. 

2754 

2755 Examples 

2756 -------- 

2757 >>> import numpy as np 

2758 >>> x = np.array([1, 2, 3, 4]) 

2759 >>> x.shape 

2760 (4,) 

2761 >>> y = np.zeros((2, 3, 4)) 

2762 >>> y.shape 

2763 (2, 3, 4) 

2764 >>> y.shape = (3, 8) 

2765 >>> y 

2766 array([[ 0., 0., 0., 0., 0., 0., 0., 0.], 

2767 [ 0., 0., 0., 0., 0., 0., 0., 0.], 

2768 [ 0., 0., 0., 0., 0., 0., 0., 0.]]) 

2769 >>> y.shape = (3, 6) 

2770 Traceback (most recent call last): 

2771 File "<stdin>", line 1, in <module> 

2772 ValueError: total size of new array must be unchanged 

2773 >>> np.zeros((4,2))[::2].shape = (-1,) 

2774 Traceback (most recent call last): 

2775 File "<stdin>", line 1, in <module> 

2776 AttributeError: Incompatible shape for in-place modification. Use 

2777 `.reshape()` to make a copy with the desired shape. 

2778 

2779 See Also 

2780 -------- 

2781 numpy.shape : Equivalent getter function. 

2782 numpy.reshape : Function similar to setting ``shape``. 

2783 ndarray.reshape : Method similar to setting ``shape``. 

2784 

2785 """)) 

2786 

2787 

2788add_newdoc('numpy._core.multiarray', 'ndarray', ('size', 

2789 """ 

2790 Number of elements in the array. 

2791 

2792 Equal to ``np.prod(a.shape)``, i.e., the product of the array's 

2793 dimensions. 

2794 

2795 Notes 

2796 ----- 

2797 `a.size` returns a standard arbitrary precision Python integer. This 

2798 may not be the case with other methods of obtaining the same value 

2799 (like the suggested ``np.prod(a.shape)``, which returns an instance 

2800 of ``np.int_``), and may be relevant if the value is used further in 

2801 calculations that may overflow a fixed size integer type. 

2802 

2803 Examples 

2804 -------- 

2805 >>> import numpy as np 

2806 >>> x = np.zeros((3, 5, 2), dtype=np.complex128) 

2807 >>> x.size 

2808 30 

2809 >>> np.prod(x.shape) 

2810 30 

2811 

2812 """)) 

2813 

2814 

2815add_newdoc('numpy._core.multiarray', 'ndarray', ('strides', 

2816 """ 

2817 Tuple of bytes to step in each dimension when traversing an array. 

2818 

2819 The byte offset of element ``(i[0], i[1], ..., i[n])`` in an array `a` 

2820 is:: 

2821 

2822 offset = sum(np.array(i) * a.strides) 

2823 

2824 A more detailed explanation of strides can be found in 

2825 :ref:`arrays.ndarray`. 

2826 

2827 .. warning:: 

2828 

2829 Setting ``arr.strides`` is discouraged and may be deprecated in the 

2830 future. `numpy.lib.stride_tricks.as_strided` should be preferred 

2831 to create a new view of the same data in a safer way. 

2832 

2833 Notes 

2834 ----- 

2835 Imagine an array of 32-bit integers (each 4 bytes):: 

2836 

2837 x = np.array([[0, 1, 2, 3, 4], 

2838 [5, 6, 7, 8, 9]], dtype=np.int32) 

2839 

2840 This array is stored in memory as 40 bytes, one after the other 

2841 (known as a contiguous block of memory). The strides of an array tell 

2842 us how many bytes we have to skip in memory to move to the next position 

2843 along a certain axis. For example, we have to skip 4 bytes (1 value) to 

2844 move to the next column, but 20 bytes (5 values) to get to the same 

2845 position in the next row. As such, the strides for the array `x` will be 

2846 ``(20, 4)``. 

2847 

2848 See Also 

2849 -------- 

2850 numpy.lib.stride_tricks.as_strided 

2851 

2852 Examples 

2853 -------- 

2854 >>> import numpy as np 

2855 >>> y = np.reshape(np.arange(2*3*4), (2,3,4)) 

2856 >>> y 

2857 array([[[ 0, 1, 2, 3], 

2858 [ 4, 5, 6, 7], 

2859 [ 8, 9, 10, 11]], 

2860 [[12, 13, 14, 15], 

2861 [16, 17, 18, 19], 

2862 [20, 21, 22, 23]]]) 

2863 >>> y.strides 

2864 (48, 16, 4) 

2865 >>> y[1,1,1] 

2866 17 

2867 >>> offset=sum(y.strides * np.array((1,1,1))) 

2868 >>> offset/y.itemsize 

2869 17 

2870 

2871 >>> x = np.reshape(np.arange(5*6*7*8), (5,6,7,8)).transpose(2,3,1,0) 

2872 >>> x.strides 

2873 (32, 4, 224, 1344) 

2874 >>> i = np.array([3,5,2,2]) 

2875 >>> offset = sum(i * x.strides) 

2876 >>> x[3,5,2,2] 

2877 813 

2878 >>> offset / x.itemsize 

2879 813 

2880 

2881 """)) 

2882 

2883 

2884add_newdoc('numpy._core.multiarray', 'ndarray', ('T', 

2885 """ 

2886 View of the transposed array. 

2887 

2888 Same as ``self.transpose()``. 

2889 

2890 Examples 

2891 -------- 

2892 >>> import numpy as np 

2893 >>> a = np.array([[1, 2], [3, 4]]) 

2894 >>> a 

2895 array([[1, 2], 

2896 [3, 4]]) 

2897 >>> a.T 

2898 array([[1, 3], 

2899 [2, 4]]) 

2900 

2901 >>> a = np.array([1, 2, 3, 4]) 

2902 >>> a 

2903 array([1, 2, 3, 4]) 

2904 >>> a.T 

2905 array([1, 2, 3, 4]) 

2906 

2907 See Also 

2908 -------- 

2909 transpose 

2910 

2911 """)) 

2912 

2913 

2914add_newdoc('numpy._core.multiarray', 'ndarray', ('mT', 

2915 """ 

2916 View of the matrix transposed array. 

2917 

2918 The matrix transpose is the transpose of the last two dimensions, even 

2919 if the array is of higher dimension. 

2920 

2921 .. versionadded:: 2.0 

2922 

2923 Raises 

2924 ------ 

2925 ValueError 

2926 If the array is of dimension less than 2. 

2927 

2928 Examples 

2929 -------- 

2930 >>> import numpy as np 

2931 >>> a = np.array([[1, 2], [3, 4]]) 

2932 >>> a 

2933 array([[1, 2], 

2934 [3, 4]]) 

2935 >>> a.mT 

2936 array([[1, 3], 

2937 [2, 4]]) 

2938 

2939 >>> a = np.arange(8).reshape((2, 2, 2)) 

2940 >>> a 

2941 array([[[0, 1], 

2942 [2, 3]], 

2943 <BLANKLINE> 

2944 [[4, 5], 

2945 [6, 7]]]) 

2946 >>> a.mT 

2947 array([[[0, 2], 

2948 [1, 3]], 

2949 <BLANKLINE> 

2950 [[4, 6], 

2951 [5, 7]]]) 

2952 

2953 """)) 

2954############################################################################## 

2955# 

2956# ndarray methods 

2957# 

2958############################################################################## 

2959 

2960 

2961add_newdoc('numpy._core.multiarray', 'ndarray', ('__array__', 

2962 """ 

2963 a.__array__([dtype], *, copy=None) 

2964 

2965 For ``dtype`` parameter it returns a new reference to self if 

2966 ``dtype`` is not given or it matches array's data type. 

2967 A new array of provided data type is returned if ``dtype`` 

2968 is different from the current data type of the array. 

2969 For ``copy`` parameter it returns a new reference to self if 

2970 ``copy=False`` or ``copy=None`` and copying isn't enforced by ``dtype`` 

2971 parameter. The method returns a new array for ``copy=True``, regardless of 

2972 ``dtype`` parameter. 

2973 

2974 A more detailed explanation of the ``__array__`` interface 

2975 can be found in :ref:`dunder_array.interface`. 

2976 

2977 """)) 

2978 

2979 

2980add_newdoc('numpy._core.multiarray', 'ndarray', ('__array_finalize__', 

2981 """ 

2982 a.__array_finalize__(obj, /) 

2983 

2984 Present so subclasses can call super. Does nothing. 

2985 

2986 """)) 

2987 

2988 

2989add_newdoc('numpy._core.multiarray', 'ndarray', ('__array_wrap__', 

2990 """ 

2991 a.__array_wrap__(array[, context], /) 

2992 

2993 Returns a view of `array` with the same type as self. 

2994 

2995 """)) 

2996 

2997 

2998add_newdoc('numpy._core.multiarray', 'ndarray', ('__copy__', 

2999 """ 

3000 a.__copy__() 

3001 

3002 Used if :func:`copy.copy` is called on an array. Returns a copy of the array. 

3003 

3004 Equivalent to ``a.copy(order='K')``. 

3005 

3006 """)) 

3007 

3008 

3009add_newdoc('numpy._core.multiarray', 'ndarray', ('__class_getitem__', 

3010 """ 

3011 a.__class_getitem__(item, /) 

3012 

3013 Return a parametrized wrapper around the `~numpy.ndarray` type. 

3014 

3015 .. versionadded:: 1.22 

3016 

3017 Returns 

3018 ------- 

3019 alias : types.GenericAlias 

3020 A parametrized `~numpy.ndarray` type. 

3021 

3022 Examples 

3023 -------- 

3024 >>> from typing import Any 

3025 >>> import numpy as np 

3026 

3027 >>> np.ndarray[Any, np.dtype[Any]] 

3028 numpy.ndarray[typing.Any, numpy.dtype[typing.Any]] 

3029 

3030 See Also 

3031 -------- 

3032 :pep:`585` : Type hinting generics in standard collections. 

3033 numpy.typing.NDArray : An ndarray alias :term:`generic <generic type>` 

3034 w.r.t. its `dtype.type <numpy.dtype.type>`. 

3035 

3036 """)) 

3037 

3038 

3039add_newdoc('numpy._core.multiarray', 'ndarray', ('__deepcopy__', 

3040 """ 

3041 a.__deepcopy__(memo, /) 

3042 

3043 Used if :func:`copy.deepcopy` is called on an array. 

3044 

3045 """)) 

3046 

3047 

3048add_newdoc('numpy._core.multiarray', 'ndarray', ('__reduce__', 

3049 """ 

3050 a.__reduce__() 

3051 

3052 For pickling. 

3053 

3054 """)) 

3055 

3056 

3057add_newdoc('numpy._core.multiarray', 'ndarray', ('__setstate__', 

3058 """ 

3059 a.__setstate__(state, /) 

3060 

3061 For unpickling. 

3062 

3063 The `state` argument must be a sequence that contains the following 

3064 elements: 

3065 

3066 Parameters 

3067 ---------- 

3068 version : int 

3069 optional pickle version. If omitted defaults to 0. 

3070 shape : tuple 

3071 dtype : data-type 

3072 isFortran : bool 

3073 rawdata : string or list 

3074 a binary string with the data (or a list if 'a' is an object array) 

3075 

3076 """)) 

3077 

3078 

3079add_newdoc('numpy._core.multiarray', 'ndarray', ('all', 

3080 """ 

3081 a.all(axis=None, out=None, keepdims=False, *, where=True) 

3082 

3083 Returns True if all elements evaluate to True. 

3084 

3085 Refer to `numpy.all` for full documentation. 

3086 

3087 See Also 

3088 -------- 

3089 numpy.all : equivalent function 

3090 

3091 """)) 

3092 

3093 

3094add_newdoc('numpy._core.multiarray', 'ndarray', ('any', 

3095 """ 

3096 a.any(axis=None, out=None, keepdims=False, *, where=True) 

3097 

3098 Returns True if any of the elements of `a` evaluate to True. 

3099 

3100 Refer to `numpy.any` for full documentation. 

3101 

3102 See Also 

3103 -------- 

3104 numpy.any : equivalent function 

3105 

3106 """)) 

3107 

3108 

3109add_newdoc('numpy._core.multiarray', 'ndarray', ('argmax', 

3110 """ 

3111 a.argmax(axis=None, out=None, *, keepdims=False) 

3112 

3113 Return indices of the maximum values along the given axis. 

3114 

3115 Refer to `numpy.argmax` for full documentation. 

3116 

3117 See Also 

3118 -------- 

3119 numpy.argmax : equivalent function 

3120 

3121 """)) 

3122 

3123 

3124add_newdoc('numpy._core.multiarray', 'ndarray', ('argmin', 

3125 """ 

3126 a.argmin(axis=None, out=None, *, keepdims=False) 

3127 

3128 Return indices of the minimum values along the given axis. 

3129 

3130 Refer to `numpy.argmin` for detailed documentation. 

3131 

3132 See Also 

3133 -------- 

3134 numpy.argmin : equivalent function 

3135 

3136 """)) 

3137 

3138 

3139add_newdoc('numpy._core.multiarray', 'ndarray', ('argsort', 

3140 """ 

3141 a.argsort(axis=-1, kind=None, order=None) 

3142 

3143 Returns the indices that would sort this array. 

3144 

3145 Refer to `numpy.argsort` for full documentation. 

3146 

3147 See Also 

3148 -------- 

3149 numpy.argsort : equivalent function 

3150 

3151 """)) 

3152 

3153 

3154add_newdoc('numpy._core.multiarray', 'ndarray', ('argpartition', 

3155 """ 

3156 a.argpartition(kth, axis=-1, kind='introselect', order=None) 

3157 

3158 Returns the indices that would partition this array. 

3159 

3160 Refer to `numpy.argpartition` for full documentation. 

3161 

3162 See Also 

3163 -------- 

3164 numpy.argpartition : equivalent function 

3165 

3166 """)) 

3167 

3168 

3169add_newdoc('numpy._core.multiarray', 'ndarray', ('astype', 

3170 """ 

3171 a.astype(dtype, order='K', casting='unsafe', subok=True, copy=True) 

3172 

3173 Copy of the array, cast to a specified type. 

3174 

3175 Parameters 

3176 ---------- 

3177 dtype : str or dtype 

3178 Typecode or data-type to which the array is cast. 

3179 order : {'C', 'F', 'A', 'K'}, optional 

3180 Controls the memory layout order of the result. 

3181 'C' means C order, 'F' means Fortran order, 'A' 

3182 means 'F' order if all the arrays are Fortran contiguous, 

3183 'C' order otherwise, and 'K' means as close to the 

3184 order the array elements appear in memory as possible. 

3185 Default is 'K'. 

3186 casting : {'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional 

3187 Controls what kind of data casting may occur. Defaults to 'unsafe' 

3188 for backwards compatibility. 

3189 

3190 * 'no' means the data types should not be cast at all. 

3191 * 'equiv' means only byte-order changes are allowed. 

3192 * 'safe' means only casts which can preserve values are allowed. 

3193 * 'same_kind' means only safe casts or casts within a kind, 

3194 like float64 to float32, are allowed. 

3195 * 'unsafe' means any data conversions may be done. 

3196 subok : bool, optional 

3197 If True, then sub-classes will be passed-through (default), otherwise 

3198 the returned array will be forced to be a base-class array. 

3199 copy : bool, optional 

3200 By default, astype always returns a newly allocated array. If this 

3201 is set to false, and the `dtype`, `order`, and `subok` 

3202 requirements are satisfied, the input array is returned instead 

3203 of a copy. 

3204 

3205 Returns 

3206 ------- 

3207 arr_t : ndarray 

3208 Unless `copy` is False and the other conditions for returning the input 

3209 array are satisfied (see description for `copy` input parameter), `arr_t` 

3210 is a new array of the same shape as the input array, with dtype, order 

3211 given by `dtype`, `order`. 

3212 

3213 Raises 

3214 ------ 

3215 ComplexWarning 

3216 When casting from complex to float or int. To avoid this, 

3217 one should use ``a.real.astype(t)``. 

3218 

3219 Examples 

3220 -------- 

3221 >>> import numpy as np 

3222 >>> x = np.array([1, 2, 2.5]) 

3223 >>> x 

3224 array([1. , 2. , 2.5]) 

3225 

3226 >>> x.astype(int) 

3227 array([1, 2, 2]) 

3228 

3229 """)) 

3230 

3231 

3232add_newdoc('numpy._core.multiarray', 'ndarray', ('byteswap', 

3233 """ 

3234 a.byteswap(inplace=False) 

3235 

3236 Swap the bytes of the array elements 

3237 

3238 Toggle between low-endian and big-endian data representation by 

3239 returning a byteswapped array, optionally swapped in-place. 

3240 Arrays of byte-strings are not swapped. The real and imaginary 

3241 parts of a complex number are swapped individually. 

3242 

3243 Parameters 

3244 ---------- 

3245 inplace : bool, optional 

3246 If ``True``, swap bytes in-place, default is ``False``. 

3247 

3248 Returns 

3249 ------- 

3250 out : ndarray 

3251 The byteswapped array. If `inplace` is ``True``, this is 

3252 a view to self. 

3253 

3254 Examples 

3255 -------- 

3256 >>> import numpy as np 

3257 >>> A = np.array([1, 256, 8755], dtype=np.int16) 

3258 >>> list(map(hex, A)) 

3259 ['0x1', '0x100', '0x2233'] 

3260 >>> A.byteswap(inplace=True) 

3261 array([ 256, 1, 13090], dtype=int16) 

3262 >>> list(map(hex, A)) 

3263 ['0x100', '0x1', '0x3322'] 

3264 

3265 Arrays of byte-strings are not swapped 

3266 

3267 >>> A = np.array([b'ceg', b'fac']) 

3268 >>> A.byteswap() 

3269 array([b'ceg', b'fac'], dtype='|S3') 

3270 

3271 ``A.view(A.dtype.newbyteorder()).byteswap()`` produces an array with 

3272 the same values but different representation in memory 

3273 

3274 >>> A = np.array([1, 2, 3],dtype=np.int64) 

3275 >>> A.view(np.uint8) 

3276 array([1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 

3277 0, 0], dtype=uint8) 

3278 >>> A.view(A.dtype.newbyteorder()).byteswap(inplace=True) 

3279 array([1, 2, 3], dtype='>i8') 

3280 >>> A.view(np.uint8) 

3281 array([0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 

3282 0, 3], dtype=uint8) 

3283 

3284 """)) 

3285 

3286 

3287add_newdoc('numpy._core.multiarray', 'ndarray', ('choose', 

3288 """ 

3289 a.choose(choices, out=None, mode='raise') 

3290 

3291 Use an index array to construct a new array from a set of choices. 

3292 

3293 Refer to `numpy.choose` for full documentation. 

3294 

3295 See Also 

3296 -------- 

3297 numpy.choose : equivalent function 

3298 

3299 """)) 

3300 

3301 

3302add_newdoc('numpy._core.multiarray', 'ndarray', ('clip', 

3303 """ 

3304 a.clip(min=None, max=None, out=None, **kwargs) 

3305 

3306 Return an array whose values are limited to ``[min, max]``. 

3307 One of max or min must be given. 

3308 

3309 Refer to `numpy.clip` for full documentation. 

3310 

3311 See Also 

3312 -------- 

3313 numpy.clip : equivalent function 

3314 

3315 """)) 

3316 

3317 

3318add_newdoc('numpy._core.multiarray', 'ndarray', ('compress', 

3319 """ 

3320 a.compress(condition, axis=None, out=None) 

3321 

3322 Return selected slices of this array along given axis. 

3323 

3324 Refer to `numpy.compress` for full documentation. 

3325 

3326 See Also 

3327 -------- 

3328 numpy.compress : equivalent function 

3329 

3330 """)) 

3331 

3332 

3333add_newdoc('numpy._core.multiarray', 'ndarray', ('conj', 

3334 """ 

3335 a.conj() 

3336 

3337 Complex-conjugate all elements. 

3338 

3339 Refer to `numpy.conjugate` for full documentation. 

3340 

3341 See Also 

3342 -------- 

3343 numpy.conjugate : equivalent function 

3344 

3345 """)) 

3346 

3347 

3348add_newdoc('numpy._core.multiarray', 'ndarray', ('conjugate', 

3349 """ 

3350 a.conjugate() 

3351 

3352 Return the complex conjugate, element-wise. 

3353 

3354 Refer to `numpy.conjugate` for full documentation. 

3355 

3356 See Also 

3357 -------- 

3358 numpy.conjugate : equivalent function 

3359 

3360 """)) 

3361 

3362 

3363add_newdoc('numpy._core.multiarray', 'ndarray', ('copy', 

3364 """ 

3365 a.copy(order='C') 

3366 

3367 Return a copy of the array. 

3368 

3369 Parameters 

3370 ---------- 

3371 order : {'C', 'F', 'A', 'K'}, optional 

3372 Controls the memory layout of the copy. 'C' means C-order, 

3373 'F' means F-order, 'A' means 'F' if `a` is Fortran contiguous, 

3374 'C' otherwise. 'K' means match the layout of `a` as closely 

3375 as possible. (Note that this function and :func:`numpy.copy` are very 

3376 similar but have different default values for their order= 

3377 arguments, and this function always passes sub-classes through.) 

3378 

3379 See also 

3380 -------- 

3381 numpy.copy : Similar function with different default behavior 

3382 numpy.copyto 

3383 

3384 Notes 

3385 ----- 

3386 This function is the preferred method for creating an array copy. The 

3387 function :func:`numpy.copy` is similar, but it defaults to using order 'K', 

3388 and will not pass sub-classes through by default. 

3389 

3390 Examples 

3391 -------- 

3392 >>> import numpy as np 

3393 >>> x = np.array([[1,2,3],[4,5,6]], order='F') 

3394 

3395 >>> y = x.copy() 

3396 

3397 >>> x.fill(0) 

3398 

3399 >>> x 

3400 array([[0, 0, 0], 

3401 [0, 0, 0]]) 

3402 

3403 >>> y 

3404 array([[1, 2, 3], 

3405 [4, 5, 6]]) 

3406 

3407 >>> y.flags['C_CONTIGUOUS'] 

3408 True 

3409 

3410 For arrays containing Python objects (e.g. dtype=object), 

3411 the copy is a shallow one. The new array will contain the 

3412 same object which may lead to surprises if that object can 

3413 be modified (is mutable): 

3414 

3415 >>> a = np.array([1, 'm', [2, 3, 4]], dtype=object) 

3416 >>> b = a.copy() 

3417 >>> b[2][0] = 10 

3418 >>> a 

3419 array([1, 'm', list([10, 3, 4])], dtype=object) 

3420 

3421 To ensure all elements within an ``object`` array are copied, 

3422 use `copy.deepcopy`: 

3423 

3424 >>> import copy 

3425 >>> a = np.array([1, 'm', [2, 3, 4]], dtype=object) 

3426 >>> c = copy.deepcopy(a) 

3427 >>> c[2][0] = 10 

3428 >>> c 

3429 array([1, 'm', list([10, 3, 4])], dtype=object) 

3430 >>> a 

3431 array([1, 'm', list([2, 3, 4])], dtype=object) 

3432 

3433 """)) 

3434 

3435 

3436add_newdoc('numpy._core.multiarray', 'ndarray', ('cumprod', 

3437 """ 

3438 a.cumprod(axis=None, dtype=None, out=None) 

3439 

3440 Return the cumulative product of the elements along the given axis. 

3441 

3442 Refer to `numpy.cumprod` for full documentation. 

3443 

3444 See Also 

3445 -------- 

3446 numpy.cumprod : equivalent function 

3447 

3448 """)) 

3449 

3450 

3451add_newdoc('numpy._core.multiarray', 'ndarray', ('cumsum', 

3452 """ 

3453 a.cumsum(axis=None, dtype=None, out=None) 

3454 

3455 Return the cumulative sum of the elements along the given axis. 

3456 

3457 Refer to `numpy.cumsum` for full documentation. 

3458 

3459 See Also 

3460 -------- 

3461 numpy.cumsum : equivalent function 

3462 

3463 """)) 

3464 

3465 

3466add_newdoc('numpy._core.multiarray', 'ndarray', ('diagonal', 

3467 """ 

3468 a.diagonal(offset=0, axis1=0, axis2=1) 

3469 

3470 Return specified diagonals. In NumPy 1.9 the returned array is a 

3471 read-only view instead of a copy as in previous NumPy versions. In 

3472 a future version the read-only restriction will be removed. 

3473 

3474 Refer to :func:`numpy.diagonal` for full documentation. 

3475 

3476 See Also 

3477 -------- 

3478 numpy.diagonal : equivalent function 

3479 

3480 """)) 

3481 

3482 

3483add_newdoc('numpy._core.multiarray', 'ndarray', ('dot')) 

3484 

3485 

3486add_newdoc('numpy._core.multiarray', 'ndarray', ('dump', 

3487 """ 

3488 a.dump(file) 

3489 

3490 Dump a pickle of the array to the specified file. 

3491 The array can be read back with pickle.load or numpy.load. 

3492 

3493 Parameters 

3494 ---------- 

3495 file : str or Path 

3496 A string naming the dump file. 

3497 

3498 """)) 

3499 

3500 

3501add_newdoc('numpy._core.multiarray', 'ndarray', ('dumps', 

3502 """ 

3503 a.dumps() 

3504 

3505 Returns the pickle of the array as a string. 

3506 pickle.loads will convert the string back to an array. 

3507 

3508 Parameters 

3509 ---------- 

3510 None 

3511 

3512 """)) 

3513 

3514 

3515add_newdoc('numpy._core.multiarray', 'ndarray', ('fill', 

3516 """ 

3517 a.fill(value) 

3518 

3519 Fill the array with a scalar value. 

3520 

3521 Parameters 

3522 ---------- 

3523 value : scalar 

3524 All elements of `a` will be assigned this value. 

3525 

3526 Examples 

3527 -------- 

3528 >>> import numpy as np 

3529 >>> a = np.array([1, 2]) 

3530 >>> a.fill(0) 

3531 >>> a 

3532 array([0, 0]) 

3533 >>> a = np.empty(2) 

3534 >>> a.fill(1) 

3535 >>> a 

3536 array([1., 1.]) 

3537 

3538 Fill expects a scalar value and always behaves the same as assigning 

3539 to a single array element. The following is a rare example where this 

3540 distinction is important: 

3541 

3542 >>> a = np.array([None, None], dtype=object) 

3543 >>> a[0] = np.array(3) 

3544 >>> a 

3545 array([array(3), None], dtype=object) 

3546 >>> a.fill(np.array(3)) 

3547 >>> a 

3548 array([array(3), array(3)], dtype=object) 

3549 

3550 Where other forms of assignments will unpack the array being assigned: 

3551 

3552 >>> a[...] = np.array(3) 

3553 >>> a 

3554 array([3, 3], dtype=object) 

3555 

3556 """)) 

3557 

3558 

3559add_newdoc('numpy._core.multiarray', 'ndarray', ('flatten', 

3560 """ 

3561 a.flatten(order='C') 

3562 

3563 Return a copy of the array collapsed into one dimension. 

3564 

3565 Parameters 

3566 ---------- 

3567 order : {'C', 'F', 'A', 'K'}, optional 

3568 'C' means to flatten in row-major (C-style) order. 

3569 'F' means to flatten in column-major (Fortran- 

3570 style) order. 'A' means to flatten in column-major 

3571 order if `a` is Fortran *contiguous* in memory, 

3572 row-major order otherwise. 'K' means to flatten 

3573 `a` in the order the elements occur in memory. 

3574 The default is 'C'. 

3575 

3576 Returns 

3577 ------- 

3578 y : ndarray 

3579 A copy of the input array, flattened to one dimension. 

3580 

3581 See Also 

3582 -------- 

3583 ravel : Return a flattened array. 

3584 flat : A 1-D flat iterator over the array. 

3585 

3586 Examples 

3587 -------- 

3588 >>> import numpy as np 

3589 >>> a = np.array([[1,2], [3,4]]) 

3590 >>> a.flatten() 

3591 array([1, 2, 3, 4]) 

3592 >>> a.flatten('F') 

3593 array([1, 3, 2, 4]) 

3594 

3595 """)) 

3596 

3597 

3598add_newdoc('numpy._core.multiarray', 'ndarray', ('getfield', 

3599 """ 

3600 a.getfield(dtype, offset=0) 

3601 

3602 Returns a field of the given array as a certain type. 

3603 

3604 A field is a view of the array data with a given data-type. The values in 

3605 the view are determined by the given type and the offset into the current 

3606 array in bytes. The offset needs to be such that the view dtype fits in the 

3607 array dtype; for example an array of dtype complex128 has 16-byte elements. 

3608 If taking a view with a 32-bit integer (4 bytes), the offset needs to be 

3609 between 0 and 12 bytes. 

3610 

3611 Parameters 

3612 ---------- 

3613 dtype : str or dtype 

3614 The data type of the view. The dtype size of the view can not be larger 

3615 than that of the array itself. 

3616 offset : int 

3617 Number of bytes to skip before beginning the element view. 

3618 

3619 Examples 

3620 -------- 

3621 >>> import numpy as np 

3622 >>> x = np.diag([1.+1.j]*2) 

3623 >>> x[1, 1] = 2 + 4.j 

3624 >>> x 

3625 array([[1.+1.j, 0.+0.j], 

3626 [0.+0.j, 2.+4.j]]) 

3627 >>> x.getfield(np.float64) 

3628 array([[1., 0.], 

3629 [0., 2.]]) 

3630 

3631 By choosing an offset of 8 bytes we can select the complex part of the 

3632 array for our view: 

3633 

3634 >>> x.getfield(np.float64, offset=8) 

3635 array([[1., 0.], 

3636 [0., 4.]]) 

3637 

3638 """)) 

3639 

3640 

3641add_newdoc('numpy._core.multiarray', 'ndarray', ('item', 

3642 """ 

3643 a.item(*args) 

3644 

3645 Copy an element of an array to a standard Python scalar and return it. 

3646 

3647 Parameters 

3648 ---------- 

3649 \\*args : Arguments (variable number and type) 

3650 

3651 * none: in this case, the method only works for arrays 

3652 with one element (`a.size == 1`), which element is 

3653 copied into a standard Python scalar object and returned. 

3654 

3655 * int_type: this argument is interpreted as a flat index into 

3656 the array, specifying which element to copy and return. 

3657 

3658 * tuple of int_types: functions as does a single int_type argument, 

3659 except that the argument is interpreted as an nd-index into the 

3660 array. 

3661 

3662 Returns 

3663 ------- 

3664 z : Standard Python scalar object 

3665 A copy of the specified element of the array as a suitable 

3666 Python scalar 

3667 

3668 Notes 

3669 ----- 

3670 When the data type of `a` is longdouble or clongdouble, item() returns 

3671 a scalar array object because there is no available Python scalar that 

3672 would not lose information. Void arrays return a buffer object for item(), 

3673 unless fields are defined, in which case a tuple is returned. 

3674 

3675 `item` is very similar to a[args], except, instead of an array scalar, 

3676 a standard Python scalar is returned. This can be useful for speeding up 

3677 access to elements of the array and doing arithmetic on elements of the 

3678 array using Python's optimized math. 

3679 

3680 Examples 

3681 -------- 

3682 >>> import numpy as np 

3683 >>> np.random.seed(123) 

3684 >>> x = np.random.randint(9, size=(3, 3)) 

3685 >>> x 

3686 array([[2, 2, 6], 

3687 [1, 3, 6], 

3688 [1, 0, 1]]) 

3689 >>> x.item(3) 

3690 1 

3691 >>> x.item(7) 

3692 0 

3693 >>> x.item((0, 1)) 

3694 2 

3695 >>> x.item((2, 2)) 

3696 1 

3697 

3698 For an array with object dtype, elements are returned as-is. 

3699 

3700 >>> a = np.array([np.int64(1)], dtype=object) 

3701 >>> a.item() #return np.int64 

3702 np.int64(1) 

3703 

3704 """)) 

3705 

3706 

3707add_newdoc('numpy._core.multiarray', 'ndarray', ('max', 

3708 """ 

3709 a.max(axis=None, out=None, keepdims=False, initial=<no value>, where=True) 

3710 

3711 Return the maximum along a given axis. 

3712 

3713 Refer to `numpy.amax` for full documentation. 

3714 

3715 See Also 

3716 -------- 

3717 numpy.amax : equivalent function 

3718 

3719 """)) 

3720 

3721 

3722add_newdoc('numpy._core.multiarray', 'ndarray', ('mean', 

3723 """ 

3724 a.mean(axis=None, dtype=None, out=None, keepdims=False, *, where=True) 

3725 

3726 Returns the average of the array elements along given axis. 

3727 

3728 Refer to `numpy.mean` for full documentation. 

3729 

3730 See Also 

3731 -------- 

3732 numpy.mean : equivalent function 

3733 

3734 """)) 

3735 

3736 

3737add_newdoc('numpy._core.multiarray', 'ndarray', ('min', 

3738 """ 

3739 a.min(axis=None, out=None, keepdims=False, initial=<no value>, where=True) 

3740 

3741 Return the minimum along a given axis. 

3742 

3743 Refer to `numpy.amin` for full documentation. 

3744 

3745 See Also 

3746 -------- 

3747 numpy.amin : equivalent function 

3748 

3749 """)) 

3750 

3751 

3752add_newdoc('numpy._core.multiarray', 'ndarray', ('nonzero', 

3753 """ 

3754 a.nonzero() 

3755 

3756 Return the indices of the elements that are non-zero. 

3757 

3758 Refer to `numpy.nonzero` for full documentation. 

3759 

3760 See Also 

3761 -------- 

3762 numpy.nonzero : equivalent function 

3763 

3764 """)) 

3765 

3766 

3767add_newdoc('numpy._core.multiarray', 'ndarray', ('prod', 

3768 """ 

3769 a.prod(axis=None, dtype=None, out=None, keepdims=False, 

3770 initial=1, where=True) 

3771 

3772 Return the product of the array elements over the given axis 

3773 

3774 Refer to `numpy.prod` for full documentation. 

3775 

3776 See Also 

3777 -------- 

3778 numpy.prod : equivalent function 

3779 

3780 """)) 

3781 

3782 

3783add_newdoc('numpy._core.multiarray', 'ndarray', ('put', 

3784 """ 

3785 a.put(indices, values, mode='raise') 

3786 

3787 Set ``a.flat[n] = values[n]`` for all `n` in indices. 

3788 

3789 Refer to `numpy.put` for full documentation. 

3790 

3791 See Also 

3792 -------- 

3793 numpy.put : equivalent function 

3794 

3795 """)) 

3796 

3797 

3798add_newdoc('numpy._core.multiarray', 'ndarray', ('ravel', 

3799 """ 

3800 a.ravel([order]) 

3801 

3802 Return a flattened array. 

3803 

3804 Refer to `numpy.ravel` for full documentation. 

3805 

3806 See Also 

3807 -------- 

3808 numpy.ravel : equivalent function 

3809 

3810 ndarray.flat : a flat iterator on the array. 

3811 

3812 """)) 

3813 

3814 

3815add_newdoc('numpy._core.multiarray', 'ndarray', ('repeat', 

3816 """ 

3817 a.repeat(repeats, axis=None) 

3818 

3819 Repeat elements of an array. 

3820 

3821 Refer to `numpy.repeat` for full documentation. 

3822 

3823 See Also 

3824 -------- 

3825 numpy.repeat : equivalent function 

3826 

3827 """)) 

3828 

3829 

3830add_newdoc('numpy._core.multiarray', 'ndarray', ('reshape', 

3831 """ 

3832 a.reshape(shape, /, *, order='C', copy=None) 

3833 

3834 Returns an array containing the same data with a new shape. 

3835 

3836 Refer to `numpy.reshape` for full documentation. 

3837 

3838 See Also 

3839 -------- 

3840 numpy.reshape : equivalent function 

3841 

3842 Notes 

3843 ----- 

3844 Unlike the free function `numpy.reshape`, this method on `ndarray` allows 

3845 the elements of the shape parameter to be passed in as separate arguments. 

3846 For example, ``a.reshape(10, 11)`` is equivalent to 

3847 ``a.reshape((10, 11))``. 

3848 

3849 """)) 

3850 

3851 

3852add_newdoc('numpy._core.multiarray', 'ndarray', ('resize', 

3853 """ 

3854 a.resize(new_shape, refcheck=True) 

3855 

3856 Change shape and size of array in-place. 

3857 

3858 Parameters 

3859 ---------- 

3860 new_shape : tuple of ints, or `n` ints 

3861 Shape of resized array. 

3862 refcheck : bool, optional 

3863 If False, reference count will not be checked. Default is True. 

3864 

3865 Returns 

3866 ------- 

3867 None 

3868 

3869 Raises 

3870 ------ 

3871 ValueError 

3872 If `a` does not own its own data or references or views to it exist, 

3873 and the data memory must be changed. 

3874 PyPy only: will always raise if the data memory must be changed, since 

3875 there is no reliable way to determine if references or views to it 

3876 exist. 

3877 

3878 SystemError 

3879 If the `order` keyword argument is specified. This behaviour is a 

3880 bug in NumPy. 

3881 

3882 See Also 

3883 -------- 

3884 resize : Return a new array with the specified shape. 

3885 

3886 Notes 

3887 ----- 

3888 This reallocates space for the data area if necessary. 

3889 

3890 Only contiguous arrays (data elements consecutive in memory) can be 

3891 resized. 

3892 

3893 The purpose of the reference count check is to make sure you 

3894 do not use this array as a buffer for another Python object and then 

3895 reallocate the memory. However, reference counts can increase in 

3896 other ways so if you are sure that you have not shared the memory 

3897 for this array with another Python object, then you may safely set 

3898 `refcheck` to False. 

3899 

3900 Examples 

3901 -------- 

3902 Shrinking an array: array is flattened (in the order that the data are 

3903 stored in memory), resized, and reshaped: 

3904 

3905 >>> import numpy as np 

3906 

3907 >>> a = np.array([[0, 1], [2, 3]], order='C') 

3908 >>> a.resize((2, 1)) 

3909 >>> a 

3910 array([[0], 

3911 [1]]) 

3912 

3913 >>> a = np.array([[0, 1], [2, 3]], order='F') 

3914 >>> a.resize((2, 1)) 

3915 >>> a 

3916 array([[0], 

3917 [2]]) 

3918 

3919 Enlarging an array: as above, but missing entries are filled with zeros: 

3920 

3921 >>> b = np.array([[0, 1], [2, 3]]) 

3922 >>> b.resize(2, 3) # new_shape parameter doesn't have to be a tuple 

3923 >>> b 

3924 array([[0, 1, 2], 

3925 [3, 0, 0]]) 

3926 

3927 Referencing an array prevents resizing... 

3928 

3929 >>> c = a 

3930 >>> a.resize((1, 1)) 

3931 Traceback (most recent call last): 

3932 ... 

3933 ValueError: cannot resize an array that references or is referenced ... 

3934 

3935 Unless `refcheck` is False: 

3936 

3937 >>> a.resize((1, 1), refcheck=False) 

3938 >>> a 

3939 array([[0]]) 

3940 >>> c 

3941 array([[0]]) 

3942 

3943 """)) 

3944 

3945 

3946add_newdoc('numpy._core.multiarray', 'ndarray', ('round', 

3947 """ 

3948 a.round(decimals=0, out=None) 

3949 

3950 Return `a` with each element rounded to the given number of decimals. 

3951 

3952 Refer to `numpy.around` for full documentation. 

3953 

3954 See Also 

3955 -------- 

3956 numpy.around : equivalent function 

3957 

3958 """)) 

3959 

3960 

3961add_newdoc('numpy._core.multiarray', 'ndarray', ('searchsorted', 

3962 """ 

3963 a.searchsorted(v, side='left', sorter=None) 

3964 

3965 Find indices where elements of v should be inserted in a to maintain order. 

3966 

3967 For full documentation, see `numpy.searchsorted` 

3968 

3969 See Also 

3970 -------- 

3971 numpy.searchsorted : equivalent function 

3972 

3973 """)) 

3974 

3975 

3976add_newdoc('numpy._core.multiarray', 'ndarray', ('setfield', 

3977 """ 

3978 a.setfield(val, dtype, offset=0) 

3979 

3980 Put a value into a specified place in a field defined by a data-type. 

3981 

3982 Place `val` into `a`'s field defined by `dtype` and beginning `offset` 

3983 bytes into the field. 

3984 

3985 Parameters 

3986 ---------- 

3987 val : object 

3988 Value to be placed in field. 

3989 dtype : dtype object 

3990 Data-type of the field in which to place `val`. 

3991 offset : int, optional 

3992 The number of bytes into the field at which to place `val`. 

3993 

3994 Returns 

3995 ------- 

3996 None 

3997 

3998 See Also 

3999 -------- 

4000 getfield 

4001 

4002 Examples 

4003 -------- 

4004 >>> import numpy as np 

4005 >>> x = np.eye(3) 

4006 >>> x.getfield(np.float64) 

4007 array([[1., 0., 0.], 

4008 [0., 1., 0.], 

4009 [0., 0., 1.]]) 

4010 >>> x.setfield(3, np.int32) 

4011 >>> x.getfield(np.int32) 

4012 array([[3, 3, 3], 

4013 [3, 3, 3], 

4014 [3, 3, 3]], dtype=int32) 

4015 >>> x 

4016 array([[1.0e+000, 1.5e-323, 1.5e-323], 

4017 [1.5e-323, 1.0e+000, 1.5e-323], 

4018 [1.5e-323, 1.5e-323, 1.0e+000]]) 

4019 >>> x.setfield(np.eye(3), np.int32) 

4020 >>> x 

4021 array([[1., 0., 0.], 

4022 [0., 1., 0.], 

4023 [0., 0., 1.]]) 

4024 

4025 """)) 

4026 

4027 

4028add_newdoc('numpy._core.multiarray', 'ndarray', ('setflags', 

4029 """ 

4030 a.setflags(write=None, align=None, uic=None) 

4031 

4032 Set array flags WRITEABLE, ALIGNED, WRITEBACKIFCOPY, 

4033 respectively. 

4034 

4035 These Boolean-valued flags affect how numpy interprets the memory 

4036 area used by `a` (see Notes below). The ALIGNED flag can only 

4037 be set to True if the data is actually aligned according to the type. 

4038 The WRITEBACKIFCOPY flag can never be set 

4039 to True. The flag WRITEABLE can only be set to True if the array owns its 

4040 own memory, or the ultimate owner of the memory exposes a writeable buffer 

4041 interface, or is a string. (The exception for string is made so that 

4042 unpickling can be done without copying memory.) 

4043 

4044 Parameters 

4045 ---------- 

4046 write : bool, optional 

4047 Describes whether or not `a` can be written to. 

4048 align : bool, optional 

4049 Describes whether or not `a` is aligned properly for its type. 

4050 uic : bool, optional 

4051 Describes whether or not `a` is a copy of another "base" array. 

4052 

4053 Notes 

4054 ----- 

4055 Array flags provide information about how the memory area used 

4056 for the array is to be interpreted. There are 7 Boolean flags 

4057 in use, only three of which can be changed by the user: 

4058 WRITEBACKIFCOPY, WRITEABLE, and ALIGNED. 

4059 

4060 WRITEABLE (W) the data area can be written to; 

4061 

4062 ALIGNED (A) the data and strides are aligned appropriately for the hardware 

4063 (as determined by the compiler); 

4064 

4065 WRITEBACKIFCOPY (X) this array is a copy of some other array (referenced 

4066 by .base). When the C-API function PyArray_ResolveWritebackIfCopy is 

4067 called, the base array will be updated with the contents of this array. 

4068 

4069 All flags can be accessed using the single (upper case) letter as well 

4070 as the full name. 

4071 

4072 Examples 

4073 -------- 

4074 >>> import numpy as np 

4075 >>> y = np.array([[3, 1, 7], 

4076 ... [2, 0, 0], 

4077 ... [8, 5, 9]]) 

4078 >>> y 

4079 array([[3, 1, 7], 

4080 [2, 0, 0], 

4081 [8, 5, 9]]) 

4082 >>> y.flags 

4083 C_CONTIGUOUS : True 

4084 F_CONTIGUOUS : False 

4085 OWNDATA : True 

4086 WRITEABLE : True 

4087 ALIGNED : True 

4088 WRITEBACKIFCOPY : False 

4089 >>> y.setflags(write=0, align=0) 

4090 >>> y.flags 

4091 C_CONTIGUOUS : True 

4092 F_CONTIGUOUS : False 

4093 OWNDATA : True 

4094 WRITEABLE : False 

4095 ALIGNED : False 

4096 WRITEBACKIFCOPY : False 

4097 >>> y.setflags(uic=1) 

4098 Traceback (most recent call last): 

4099 File "<stdin>", line 1, in <module> 

4100 ValueError: cannot set WRITEBACKIFCOPY flag to True 

4101 

4102 """)) 

4103 

4104 

4105add_newdoc('numpy._core.multiarray', 'ndarray', ('sort', 

4106 """ 

4107 a.sort(axis=-1, kind=None, order=None) 

4108 

4109 Sort an array in-place. Refer to `numpy.sort` for full documentation. 

4110 

4111 Parameters 

4112 ---------- 

4113 axis : int, optional 

4114 Axis along which to sort. Default is -1, which means sort along the 

4115 last axis. 

4116 kind : {'quicksort', 'mergesort', 'heapsort', 'stable'}, optional 

4117 Sorting algorithm. The default is 'quicksort'. Note that both 'stable' 

4118 and 'mergesort' use timsort under the covers and, in general, the 

4119 actual implementation will vary with datatype. The 'mergesort' option 

4120 is retained for backwards compatibility. 

4121 order : str or list of str, optional 

4122 When `a` is an array with fields defined, this argument specifies 

4123 which fields to compare first, second, etc. A single field can 

4124 be specified as a string, and not all fields need be specified, 

4125 but unspecified fields will still be used, in the order in which 

4126 they come up in the dtype, to break ties. 

4127 

4128 See Also 

4129 -------- 

4130 numpy.sort : Return a sorted copy of an array. 

4131 numpy.argsort : Indirect sort. 

4132 numpy.lexsort : Indirect stable sort on multiple keys. 

4133 numpy.searchsorted : Find elements in sorted array. 

4134 numpy.partition: Partial sort. 

4135 

4136 Notes 

4137 ----- 

4138 See `numpy.sort` for notes on the different sorting algorithms. 

4139 

4140 Examples 

4141 -------- 

4142 >>> import numpy as np 

4143 >>> a = np.array([[1,4], [3,1]]) 

4144 >>> a.sort(axis=1) 

4145 >>> a 

4146 array([[1, 4], 

4147 [1, 3]]) 

4148 >>> a.sort(axis=0) 

4149 >>> a 

4150 array([[1, 3], 

4151 [1, 4]]) 

4152 

4153 Use the `order` keyword to specify a field to use when sorting a 

4154 structured array: 

4155 

4156 >>> a = np.array([('a', 2), ('c', 1)], dtype=[('x', 'S1'), ('y', int)]) 

4157 >>> a.sort(order='y') 

4158 >>> a 

4159 array([(b'c', 1), (b'a', 2)], 

4160 dtype=[('x', 'S1'), ('y', '<i8')]) 

4161 

4162 """)) 

4163 

4164 

4165add_newdoc('numpy._core.multiarray', 'ndarray', ('partition', 

4166 """ 

4167 a.partition(kth, axis=-1, kind='introselect', order=None) 

4168 

4169 Partially sorts the elements in the array in such a way that the value of 

4170 the element in k-th position is in the position it would be in a sorted 

4171 array. In the output array, all elements smaller than the k-th element 

4172 are located to the left of this element and all equal or greater are 

4173 located to its right. The ordering of the elements in the two partitions 

4174 on the either side of the k-th element in the output array is undefined. 

4175 

4176 Parameters 

4177 ---------- 

4178 kth : int or sequence of ints 

4179 Element index to partition by. The kth element value will be in its 

4180 final sorted position and all smaller elements will be moved before it 

4181 and all equal or greater elements behind it. 

4182 The order of all elements in the partitions is undefined. 

4183 If provided with a sequence of kth it will partition all elements 

4184 indexed by kth of them into their sorted position at once. 

4185 

4186 .. deprecated:: 1.22.0 

4187 Passing booleans as index is deprecated. 

4188 axis : int, optional 

4189 Axis along which to sort. Default is -1, which means sort along the 

4190 last axis. 

4191 kind : {'introselect'}, optional 

4192 Selection algorithm. Default is 'introselect'. 

4193 order : str or list of str, optional 

4194 When `a` is an array with fields defined, this argument specifies 

4195 which fields to compare first, second, etc. A single field can 

4196 be specified as a string, and not all fields need to be specified, 

4197 but unspecified fields will still be used, in the order in which 

4198 they come up in the dtype, to break ties. 

4199 

4200 See Also 

4201 -------- 

4202 numpy.partition : Return a partitioned copy of an array. 

4203 argpartition : Indirect partition. 

4204 sort : Full sort. 

4205 

4206 Notes 

4207 ----- 

4208 See ``np.partition`` for notes on the different algorithms. 

4209 

4210 Examples 

4211 -------- 

4212 >>> import numpy as np 

4213 >>> a = np.array([3, 4, 2, 1]) 

4214 >>> a.partition(3) 

4215 >>> a 

4216 array([2, 1, 3, 4]) # may vary 

4217 

4218 >>> a.partition((1, 3)) 

4219 >>> a 

4220 array([1, 2, 3, 4]) 

4221 """)) 

4222 

4223 

4224add_newdoc('numpy._core.multiarray', 'ndarray', ('squeeze', 

4225 """ 

4226 a.squeeze(axis=None) 

4227 

4228 Remove axes of length one from `a`. 

4229 

4230 Refer to `numpy.squeeze` for full documentation. 

4231 

4232 See Also 

4233 -------- 

4234 numpy.squeeze : equivalent function 

4235 

4236 """)) 

4237 

4238 

4239add_newdoc('numpy._core.multiarray', 'ndarray', ('std', 

4240 """ 

4241 a.std(axis=None, dtype=None, out=None, ddof=0, keepdims=False, *, where=True) 

4242 

4243 Returns the standard deviation of the array elements along given axis. 

4244 

4245 Refer to `numpy.std` for full documentation. 

4246 

4247 See Also 

4248 -------- 

4249 numpy.std : equivalent function 

4250 

4251 """)) 

4252 

4253 

4254add_newdoc('numpy._core.multiarray', 'ndarray', ('sum', 

4255 """ 

4256 a.sum(axis=None, dtype=None, out=None, keepdims=False, initial=0, where=True) 

4257 

4258 Return the sum of the array elements over the given axis. 

4259 

4260 Refer to `numpy.sum` for full documentation. 

4261 

4262 See Also 

4263 -------- 

4264 numpy.sum : equivalent function 

4265 

4266 """)) 

4267 

4268 

4269add_newdoc('numpy._core.multiarray', 'ndarray', ('swapaxes', 

4270 """ 

4271 a.swapaxes(axis1, axis2) 

4272 

4273 Return a view of the array with `axis1` and `axis2` interchanged. 

4274 

4275 Refer to `numpy.swapaxes` for full documentation. 

4276 

4277 See Also 

4278 -------- 

4279 numpy.swapaxes : equivalent function 

4280 

4281 """)) 

4282 

4283 

4284add_newdoc('numpy._core.multiarray', 'ndarray', ('take', 

4285 """ 

4286 a.take(indices, axis=None, out=None, mode='raise') 

4287 

4288 Return an array formed from the elements of `a` at the given indices. 

4289 

4290 Refer to `numpy.take` for full documentation. 

4291 

4292 See Also 

4293 -------- 

4294 numpy.take : equivalent function 

4295 

4296 """)) 

4297 

4298 

4299add_newdoc('numpy._core.multiarray', 'ndarray', ('tofile', 

4300 """ 

4301 a.tofile(fid, sep="", format="%s") 

4302 

4303 Write array to a file as text or binary (default). 

4304 

4305 Data is always written in 'C' order, independent of the order of `a`. 

4306 The data produced by this method can be recovered using the function 

4307 fromfile(). 

4308 

4309 Parameters 

4310 ---------- 

4311 fid : file or str or Path 

4312 An open file object, or a string containing a filename. 

4313 sep : str 

4314 Separator between array items for text output. 

4315 If "" (empty), a binary file is written, equivalent to 

4316 ``file.write(a.tobytes())``. 

4317 format : str 

4318 Format string for text file output. 

4319 Each entry in the array is formatted to text by first converting 

4320 it to the closest Python type, and then using "format" % item. 

4321 

4322 Notes 

4323 ----- 

4324 This is a convenience function for quick storage of array data. 

4325 Information on endianness and precision is lost, so this method is not a 

4326 good choice for files intended to archive data or transport data between 

4327 machines with different endianness. Some of these problems can be overcome 

4328 by outputting the data as text files, at the expense of speed and file 

4329 size. 

4330 

4331 When fid is a file object, array contents are directly written to the 

4332 file, bypassing the file object's ``write`` method. As a result, tofile 

4333 cannot be used with files objects supporting compression (e.g., GzipFile) 

4334 or file-like objects that do not support ``fileno()`` (e.g., BytesIO). 

4335 

4336 """)) 

4337 

4338 

4339add_newdoc('numpy._core.multiarray', 'ndarray', ('tolist', 

4340 """ 

4341 a.tolist() 

4342 

4343 Return the array as an ``a.ndim``-levels deep nested list of Python scalars. 

4344 

4345 Return a copy of the array data as a (nested) Python list. 

4346 Data items are converted to the nearest compatible builtin Python type, via 

4347 the `~numpy.ndarray.item` function. 

4348 

4349 If ``a.ndim`` is 0, then since the depth of the nested list is 0, it will 

4350 not be a list at all, but a simple Python scalar. 

4351 

4352 Parameters 

4353 ---------- 

4354 none 

4355 

4356 Returns 

4357 ------- 

4358 y : object, or list of object, or list of list of object, or ... 

4359 The possibly nested list of array elements. 

4360 

4361 Notes 

4362 ----- 

4363 The array may be recreated via ``a = np.array(a.tolist())``, although this 

4364 may sometimes lose precision. 

4365 

4366 Examples 

4367 -------- 

4368 For a 1D array, ``a.tolist()`` is almost the same as ``list(a)``, 

4369 except that ``tolist`` changes numpy scalars to Python scalars: 

4370 

4371 >>> import numpy as np 

4372 >>> a = np.uint32([1, 2]) 

4373 >>> a_list = list(a) 

4374 >>> a_list 

4375 [np.uint32(1), np.uint32(2)] 

4376 >>> type(a_list[0]) 

4377 <class 'numpy.uint32'> 

4378 >>> a_tolist = a.tolist() 

4379 >>> a_tolist 

4380 [1, 2] 

4381 >>> type(a_tolist[0]) 

4382 <class 'int'> 

4383 

4384 Additionally, for a 2D array, ``tolist`` applies recursively: 

4385 

4386 >>> a = np.array([[1, 2], [3, 4]]) 

4387 >>> list(a) 

4388 [array([1, 2]), array([3, 4])] 

4389 >>> a.tolist() 

4390 [[1, 2], [3, 4]] 

4391 

4392 The base case for this recursion is a 0D array: 

4393 

4394 >>> a = np.array(1) 

4395 >>> list(a) 

4396 Traceback (most recent call last): 

4397 ... 

4398 TypeError: iteration over a 0-d array 

4399 >>> a.tolist() 

4400 1 

4401 """)) 

4402 

4403 

4404add_newdoc('numpy._core.multiarray', 'ndarray', ('tobytes', """ 

4405 a.tobytes(order='C') 

4406 

4407 Construct Python bytes containing the raw data bytes in the array. 

4408 

4409 Constructs Python bytes showing a copy of the raw contents of 

4410 data memory. The bytes object is produced in C-order by default. 

4411 This behavior is controlled by the ``order`` parameter. 

4412 

4413 Parameters 

4414 ---------- 

4415 order : {'C', 'F', 'A'}, optional 

4416 Controls the memory layout of the bytes object. 'C' means C-order, 

4417 'F' means F-order, 'A' (short for *Any*) means 'F' if `a` is 

4418 Fortran contiguous, 'C' otherwise. Default is 'C'. 

4419 

4420 Returns 

4421 ------- 

4422 s : bytes 

4423 Python bytes exhibiting a copy of `a`'s raw data. 

4424 

4425 See also 

4426 -------- 

4427 frombuffer 

4428 Inverse of this operation, construct a 1-dimensional array from Python 

4429 bytes. 

4430 

4431 Examples 

4432 -------- 

4433 >>> import numpy as np 

4434 >>> x = np.array([[0, 1], [2, 3]], dtype='<u2') 

4435 >>> x.tobytes() 

4436 b'\\x00\\x00\\x01\\x00\\x02\\x00\\x03\\x00' 

4437 >>> x.tobytes('C') == x.tobytes() 

4438 True 

4439 >>> x.tobytes('F') 

4440 b'\\x00\\x00\\x02\\x00\\x01\\x00\\x03\\x00' 

4441 

4442 """)) 

4443 

4444 

4445add_newdoc('numpy._core.multiarray', 'ndarray', ('tostring', r""" 

4446 a.tostring(order='C') 

4447 

4448 A compatibility alias for `~ndarray.tobytes`, with exactly the same 

4449 behavior. 

4450 

4451 Despite its name, it returns :class:`bytes` not :class:`str`\ s. 

4452 

4453 .. deprecated:: 1.19.0 

4454 """)) 

4455 

4456 

4457add_newdoc('numpy._core.multiarray', 'ndarray', ('trace', 

4458 """ 

4459 a.trace(offset=0, axis1=0, axis2=1, dtype=None, out=None) 

4460 

4461 Return the sum along diagonals of the array. 

4462 

4463 Refer to `numpy.trace` for full documentation. 

4464 

4465 See Also 

4466 -------- 

4467 numpy.trace : equivalent function 

4468 

4469 """)) 

4470 

4471 

4472add_newdoc('numpy._core.multiarray', 'ndarray', ('transpose', 

4473 """ 

4474 a.transpose(*axes) 

4475 

4476 Returns a view of the array with axes transposed. 

4477 

4478 Refer to `numpy.transpose` for full documentation. 

4479 

4480 Parameters 

4481 ---------- 

4482 axes : None, tuple of ints, or `n` ints 

4483 

4484 * None or no argument: reverses the order of the axes. 

4485 

4486 * tuple of ints: `i` in the `j`-th place in the tuple means that the 

4487 array's `i`-th axis becomes the transposed array's `j`-th axis. 

4488 

4489 * `n` ints: same as an n-tuple of the same ints (this form is 

4490 intended simply as a "convenience" alternative to the tuple form). 

4491 

4492 Returns 

4493 ------- 

4494 p : ndarray 

4495 View of the array with its axes suitably permuted. 

4496 

4497 See Also 

4498 -------- 

4499 transpose : Equivalent function. 

4500 ndarray.T : Array property returning the array transposed. 

4501 ndarray.reshape : Give a new shape to an array without changing its data. 

4502 

4503 Examples 

4504 -------- 

4505 >>> import numpy as np 

4506 >>> a = np.array([[1, 2], [3, 4]]) 

4507 >>> a 

4508 array([[1, 2], 

4509 [3, 4]]) 

4510 >>> a.transpose() 

4511 array([[1, 3], 

4512 [2, 4]]) 

4513 >>> a.transpose((1, 0)) 

4514 array([[1, 3], 

4515 [2, 4]]) 

4516 >>> a.transpose(1, 0) 

4517 array([[1, 3], 

4518 [2, 4]]) 

4519 

4520 >>> a = np.array([1, 2, 3, 4]) 

4521 >>> a 

4522 array([1, 2, 3, 4]) 

4523 >>> a.transpose() 

4524 array([1, 2, 3, 4]) 

4525 

4526 """)) 

4527 

4528 

4529add_newdoc('numpy._core.multiarray', 'ndarray', ('var', 

4530 """ 

4531 a.var(axis=None, dtype=None, out=None, ddof=0, keepdims=False, *, where=True) 

4532 

4533 Returns the variance of the array elements, along given axis. 

4534 

4535 Refer to `numpy.var` for full documentation. 

4536 

4537 See Also 

4538 -------- 

4539 numpy.var : equivalent function 

4540 

4541 """)) 

4542 

4543 

4544add_newdoc('numpy._core.multiarray', 'ndarray', ('view', 

4545 """ 

4546 a.view([dtype][, type]) 

4547 

4548 New view of array with the same data. 

4549 

4550 .. note:: 

4551 Passing None for ``dtype`` is different from omitting the parameter, 

4552 since the former invokes ``dtype(None)`` which is an alias for 

4553 ``dtype('float64')``. 

4554 

4555 Parameters 

4556 ---------- 

4557 dtype : data-type or ndarray sub-class, optional 

4558 Data-type descriptor of the returned view, e.g., float32 or int16. 

4559 Omitting it results in the view having the same data-type as `a`. 

4560 This argument can also be specified as an ndarray sub-class, which 

4561 then specifies the type of the returned object (this is equivalent to 

4562 setting the ``type`` parameter). 

4563 type : Python type, optional 

4564 Type of the returned view, e.g., ndarray or matrix. Again, omission 

4565 of the parameter results in type preservation. 

4566 

4567 Notes 

4568 ----- 

4569 ``a.view()`` is used two different ways: 

4570 

4571 ``a.view(some_dtype)`` or ``a.view(dtype=some_dtype)`` constructs a view 

4572 of the array's memory with a different data-type. This can cause a 

4573 reinterpretation of the bytes of memory. 

4574 

4575 ``a.view(ndarray_subclass)`` or ``a.view(type=ndarray_subclass)`` just 

4576 returns an instance of `ndarray_subclass` that looks at the same array 

4577 (same shape, dtype, etc.) This does not cause a reinterpretation of the 

4578 memory. 

4579 

4580 For ``a.view(some_dtype)``, if ``some_dtype`` has a different number of 

4581 bytes per entry than the previous dtype (for example, converting a regular 

4582 array to a structured array), then the last axis of ``a`` must be 

4583 contiguous. This axis will be resized in the result. 

4584 

4585 .. versionchanged:: 1.23.0 

4586 Only the last axis needs to be contiguous. Previously, the entire array 

4587 had to be C-contiguous. 

4588 

4589 Examples 

4590 -------- 

4591 >>> import numpy as np 

4592 >>> x = np.array([(-1, 2)], dtype=[('a', np.int8), ('b', np.int8)]) 

4593 

4594 Viewing array data using a different type and dtype: 

4595 

4596 >>> nonneg = np.dtype([("a", np.uint8), ("b", np.uint8)]) 

4597 >>> y = x.view(dtype=nonneg, type=np.recarray) 

4598 >>> x["a"] 

4599 array([-1], dtype=int8) 

4600 >>> y.a 

4601 array([255], dtype=uint8) 

4602 

4603 Creating a view on a structured array so it can be used in calculations 

4604 

4605 >>> x = np.array([(1, 2),(3,4)], dtype=[('a', np.int8), ('b', np.int8)]) 

4606 >>> xv = x.view(dtype=np.int8).reshape(-1,2) 

4607 >>> xv 

4608 array([[1, 2], 

4609 [3, 4]], dtype=int8) 

4610 >>> xv.mean(0) 

4611 array([2., 3.]) 

4612 

4613 Making changes to the view changes the underlying array 

4614 

4615 >>> xv[0,1] = 20 

4616 >>> x 

4617 array([(1, 20), (3, 4)], dtype=[('a', 'i1'), ('b', 'i1')]) 

4618 

4619 Using a view to convert an array to a recarray: 

4620 

4621 >>> z = x.view(np.recarray) 

4622 >>> z.a 

4623 array([1, 3], dtype=int8) 

4624 

4625 Views share data: 

4626 

4627 >>> x[0] = (9, 10) 

4628 >>> z[0] 

4629 np.record((9, 10), dtype=[('a', 'i1'), ('b', 'i1')]) 

4630 

4631 Views that change the dtype size (bytes per entry) should normally be 

4632 avoided on arrays defined by slices, transposes, fortran-ordering, etc.: 

4633 

4634 >>> x = np.array([[1, 2, 3], [4, 5, 6]], dtype=np.int16) 

4635 >>> y = x[:, ::2] 

4636 >>> y 

4637 array([[1, 3], 

4638 [4, 6]], dtype=int16) 

4639 >>> y.view(dtype=[('width', np.int16), ('length', np.int16)]) 

4640 Traceback (most recent call last): 

4641 ... 

4642 ValueError: To change to a dtype of a different size, the last axis must be contiguous 

4643 >>> z = y.copy() 

4644 >>> z.view(dtype=[('width', np.int16), ('length', np.int16)]) 

4645 array([[(1, 3)], 

4646 [(4, 6)]], dtype=[('width', '<i2'), ('length', '<i2')]) 

4647 

4648 However, views that change dtype are totally fine for arrays with a 

4649 contiguous last axis, even if the rest of the axes are not C-contiguous: 

4650 

4651 >>> x = np.arange(2 * 3 * 4, dtype=np.int8).reshape(2, 3, 4) 

4652 >>> x.transpose(1, 0, 2).view(np.int16) 

4653 array([[[ 256, 770], 

4654 [3340, 3854]], 

4655 <BLANKLINE> 

4656 [[1284, 1798], 

4657 [4368, 4882]], 

4658 <BLANKLINE> 

4659 [[2312, 2826], 

4660 [5396, 5910]]], dtype=int16) 

4661 

4662 """)) 

4663 

4664 

4665############################################################################## 

4666# 

4667# umath functions 

4668# 

4669############################################################################## 

4670 

4671add_newdoc('numpy._core.umath', 'frompyfunc', 

4672 """ 

4673 frompyfunc(func, /, nin, nout, *[, identity]) 

4674 

4675 Takes an arbitrary Python function and returns a NumPy ufunc. 

4676 

4677 Can be used, for example, to add broadcasting to a built-in Python 

4678 function (see Examples section). 

4679 

4680 Parameters 

4681 ---------- 

4682 func : Python function object 

4683 An arbitrary Python function. 

4684 nin : int 

4685 The number of input arguments. 

4686 nout : int 

4687 The number of objects returned by `func`. 

4688 identity : object, optional 

4689 The value to use for the `~numpy.ufunc.identity` attribute of the resulting 

4690 object. If specified, this is equivalent to setting the underlying 

4691 C ``identity`` field to ``PyUFunc_IdentityValue``. 

4692 If omitted, the identity is set to ``PyUFunc_None``. Note that this is 

4693 _not_ equivalent to setting the identity to ``None``, which implies the 

4694 operation is reorderable. 

4695 

4696 Returns 

4697 ------- 

4698 out : ufunc 

4699 Returns a NumPy universal function (``ufunc``) object. 

4700 

4701 See Also 

4702 -------- 

4703 vectorize : Evaluates pyfunc over input arrays using broadcasting rules of numpy. 

4704 

4705 Notes 

4706 ----- 

4707 The returned ufunc always returns PyObject arrays. 

4708 

4709 Examples 

4710 -------- 

4711 Use frompyfunc to add broadcasting to the Python function ``oct``: 

4712 

4713 >>> import numpy as np 

4714 >>> oct_array = np.frompyfunc(oct, 1, 1) 

4715 >>> oct_array(np.array((10, 30, 100))) 

4716 array(['0o12', '0o36', '0o144'], dtype=object) 

4717 >>> np.array((oct(10), oct(30), oct(100))) # for comparison 

4718 array(['0o12', '0o36', '0o144'], dtype='<U5') 

4719 

4720 """) 

4721 

4722 

4723############################################################################## 

4724# 

4725# compiled_base functions 

4726# 

4727############################################################################## 

4728 

4729add_newdoc('numpy._core.multiarray', 'add_docstring', 

4730 """ 

4731 add_docstring(obj, docstring) 

4732 

4733 Add a docstring to a built-in obj if possible. 

4734 If the obj already has a docstring raise a RuntimeError 

4735 If this routine does not know how to add a docstring to the object 

4736 raise a TypeError 

4737 """) 

4738 

4739add_newdoc('numpy._core.umath', '_add_newdoc_ufunc', 

4740 """ 

4741 add_ufunc_docstring(ufunc, new_docstring) 

4742 

4743 Replace the docstring for a ufunc with new_docstring. 

4744 This method will only work if the current docstring for 

4745 the ufunc is NULL. (At the C level, i.e. when ufunc->doc is NULL.) 

4746 

4747 Parameters 

4748 ---------- 

4749 ufunc : numpy.ufunc 

4750 A ufunc whose current doc is NULL. 

4751 new_docstring : string 

4752 The new docstring for the ufunc. 

4753 

4754 Notes 

4755 ----- 

4756 This method allocates memory for new_docstring on 

4757 the heap. Technically this creates a memory leak, since this 

4758 memory will not be reclaimed until the end of the program 

4759 even if the ufunc itself is removed. However this will only 

4760 be a problem if the user is repeatedly creating ufuncs with 

4761 no documentation, adding documentation via add_newdoc_ufunc, 

4762 and then throwing away the ufunc. 

4763 """) 

4764 

4765add_newdoc('numpy._core.multiarray', 'get_handler_name', 

4766 """ 

4767 get_handler_name(a: ndarray) -> str,None 

4768 

4769 Return the name of the memory handler used by `a`. If not provided, return 

4770 the name of the memory handler that will be used to allocate data for the 

4771 next `ndarray` in this context. May return None if `a` does not own its 

4772 memory, in which case you can traverse ``a.base`` for a memory handler. 

4773 """) 

4774 

4775add_newdoc('numpy._core.multiarray', 'get_handler_version', 

4776 """ 

4777 get_handler_version(a: ndarray) -> int,None 

4778 

4779 Return the version of the memory handler used by `a`. If not provided, 

4780 return the version of the memory handler that will be used to allocate data 

4781 for the next `ndarray` in this context. May return None if `a` does not own 

4782 its memory, in which case you can traverse ``a.base`` for a memory handler. 

4783 """) 

4784 

4785add_newdoc('numpy._core._multiarray_umath', '_array_converter', 

4786 """ 

4787 _array_converter(*array_likes) 

4788 

4789 Helper to convert one or more objects to arrays. Integrates machinery 

4790 to deal with the ``result_type`` and ``__array_wrap__``. 

4791 

4792 The reason for this is that e.g. ``result_type`` needs to convert to arrays 

4793 to find the ``dtype``. But converting to an array before calling 

4794 ``result_type`` would incorrectly "forget" whether it was a Python int, 

4795 float, or complex. 

4796 """) 

4797 

4798add_newdoc( 

4799 'numpy._core._multiarray_umath', '_array_converter', ('scalar_input', 

4800 """ 

4801 A tuple which indicates for each input whether it was a scalar that 

4802 was coerced to a 0-D array (and was not already an array or something 

4803 converted via a protocol like ``__array__()``). 

4804 """)) 

4805 

4806add_newdoc('numpy._core._multiarray_umath', '_array_converter', ('as_arrays', 

4807 """ 

4808 as_arrays(/, subok=True, pyscalars="convert_if_no_array") 

4809 

4810 Return the inputs as arrays or scalars. 

4811 

4812 Parameters 

4813 ---------- 

4814 subok : True or False, optional 

4815 Whether array subclasses are preserved. 

4816 pyscalars : {"convert", "preserve", "convert_if_no_array"}, optional 

4817 To allow NEP 50 weak promotion later, it may be desirable to preserve 

4818 Python scalars. As default, these are preserved unless all inputs 

4819 are Python scalars. "convert" enforces an array return. 

4820 """)) 

4821 

4822add_newdoc('numpy._core._multiarray_umath', '_array_converter', ('result_type', 

4823 """result_type(/, extra_dtype=None, ensure_inexact=False) 

4824 

4825 Find the ``result_type`` just as ``np.result_type`` would, but taking 

4826 into account that the original inputs (before converting to an array) may 

4827 have been Python scalars with weak promotion. 

4828 

4829 Parameters 

4830 ---------- 

4831 extra_dtype : dtype instance or class 

4832 An additional DType or dtype instance to promote (e.g. could be used 

4833 to ensure the result precision is at least float32). 

4834 ensure_inexact : True or False 

4835 When ``True``, ensures a floating point (or complex) result replacing 

4836 the ``arr * 1.`` or ``result_type(..., 0.0)`` pattern. 

4837 """)) 

4838 

4839add_newdoc('numpy._core._multiarray_umath', '_array_converter', ('wrap', 

4840 """ 

4841 wrap(arr, /, to_scalar=None) 

4842 

4843 Call ``__array_wrap__`` on ``arr`` if ``arr`` is not the same subclass 

4844 as the input the ``__array_wrap__`` method was retrieved from. 

4845 

4846 Parameters 

4847 ---------- 

4848 arr : ndarray 

4849 The object to be wrapped. Normally an ndarray or subclass, 

4850 although for backward compatibility NumPy scalars are also accepted 

4851 (these will be converted to a NumPy array before being passed on to 

4852 the ``__array_wrap__`` method). 

4853 to_scalar : {True, False, None}, optional 

4854 When ``True`` will convert a 0-d array to a scalar via ``result[()]`` 

4855 (with a fast-path for non-subclasses). If ``False`` the result should 

4856 be an array-like (as ``__array_wrap__`` is free to return a non-array). 

4857 By default (``None``), a scalar is returned if all inputs were scalar. 

4858 """)) 

4859 

4860 

4861add_newdoc('numpy._core.multiarray', '_get_madvise_hugepage', 

4862 """ 

4863 _get_madvise_hugepage() -> bool 

4864 

4865 Get use of ``madvise (2)`` MADV_HUGEPAGE support when 

4866 allocating the array data. Returns the currently set value. 

4867 See `global_state` for more information. 

4868 """) 

4869 

4870add_newdoc('numpy._core.multiarray', '_set_madvise_hugepage', 

4871 """ 

4872 _set_madvise_hugepage(enabled: bool) -> bool 

4873 

4874 Set or unset use of ``madvise (2)`` MADV_HUGEPAGE support when 

4875 allocating the array data. Returns the previously set value. 

4876 See `global_state` for more information. 

4877 """) 

4878 

4879 

4880############################################################################## 

4881# 

4882# Documentation for ufunc attributes and methods 

4883# 

4884############################################################################## 

4885 

4886 

4887############################################################################## 

4888# 

4889# ufunc object 

4890# 

4891############################################################################## 

4892 

4893add_newdoc('numpy._core', 'ufunc', 

4894 """ 

4895 Functions that operate element by element on whole arrays. 

4896 

4897 To see the documentation for a specific ufunc, use `info`. For 

4898 example, ``np.info(np.sin)``. Because ufuncs are written in C 

4899 (for speed) and linked into Python with NumPy's ufunc facility, 

4900 Python's help() function finds this page whenever help() is called 

4901 on a ufunc. 

4902 

4903 A detailed explanation of ufuncs can be found in the docs for :ref:`ufuncs`. 

4904 

4905 **Calling ufuncs:** ``op(*x[, out], where=True, **kwargs)`` 

4906 

4907 Apply `op` to the arguments `*x` elementwise, broadcasting the arguments. 

4908 

4909 The broadcasting rules are: 

4910 

4911 * Dimensions of length 1 may be prepended to either array. 

4912 * Arrays may be repeated along dimensions of length 1. 

4913 

4914 Parameters 

4915 ---------- 

4916 *x : array_like 

4917 Input arrays. 

4918 out : ndarray, None, or tuple of ndarray and None, optional 

4919 Alternate array object(s) in which to put the result; if provided, it 

4920 must have a shape that the inputs broadcast to. A tuple of arrays 

4921 (possible only as a keyword argument) must have length equal to the 

4922 number of outputs; use None for uninitialized outputs to be 

4923 allocated by the ufunc. 

4924 where : array_like, optional 

4925 This condition is broadcast over the input. At locations where the 

4926 condition is True, the `out` array will be set to the ufunc result. 

4927 Elsewhere, the `out` array will retain its original value. 

4928 Note that if an uninitialized `out` array is created via the default 

4929 ``out=None``, locations within it where the condition is False will 

4930 remain uninitialized. 

4931 **kwargs 

4932 For other keyword-only arguments, see the :ref:`ufunc docs <ufuncs.kwargs>`. 

4933 

4934 Returns 

4935 ------- 

4936 r : ndarray or tuple of ndarray 

4937 `r` will have the shape that the arrays in `x` broadcast to; if `out` is 

4938 provided, it will be returned. If not, `r` will be allocated and 

4939 may contain uninitialized values. If the function has more than one 

4940 output, then the result will be a tuple of arrays. 

4941 

4942 """) 

4943 

4944 

4945############################################################################## 

4946# 

4947# ufunc attributes 

4948# 

4949############################################################################## 

4950 

4951add_newdoc('numpy._core', 'ufunc', ('identity', 

4952 """ 

4953 The identity value. 

4954 

4955 Data attribute containing the identity element for the ufunc, 

4956 if it has one. If it does not, the attribute value is None. 

4957 

4958 Examples 

4959 -------- 

4960 >>> import numpy as np 

4961 >>> np.add.identity 

4962 0 

4963 >>> np.multiply.identity 

4964 1 

4965 >>> np.power.identity 

4966 1 

4967 >>> print(np.exp.identity) 

4968 None 

4969 """)) 

4970 

4971add_newdoc('numpy._core', 'ufunc', ('nargs', 

4972 """ 

4973 The number of arguments. 

4974 

4975 Data attribute containing the number of arguments the ufunc takes, including 

4976 optional ones. 

4977 

4978 Notes 

4979 ----- 

4980 Typically this value will be one more than what you might expect 

4981 because all ufuncs take the optional "out" argument. 

4982 

4983 Examples 

4984 -------- 

4985 >>> import numpy as np 

4986 >>> np.add.nargs 

4987 3 

4988 >>> np.multiply.nargs 

4989 3 

4990 >>> np.power.nargs 

4991 3 

4992 >>> np.exp.nargs 

4993 2 

4994 """)) 

4995 

4996add_newdoc('numpy._core', 'ufunc', ('nin', 

4997 """ 

4998 The number of inputs. 

4999 

5000 Data attribute containing the number of arguments the ufunc treats as input. 

5001 

5002 Examples 

5003 -------- 

5004 >>> import numpy as np 

5005 >>> np.add.nin 

5006 2 

5007 >>> np.multiply.nin 

5008 2 

5009 >>> np.power.nin 

5010 2 

5011 >>> np.exp.nin 

5012 1 

5013 """)) 

5014 

5015add_newdoc('numpy._core', 'ufunc', ('nout', 

5016 """ 

5017 The number of outputs. 

5018 

5019 Data attribute containing the number of arguments the ufunc treats as output. 

5020 

5021 Notes 

5022 ----- 

5023 Since all ufuncs can take output arguments, this will always be at least 1. 

5024 

5025 Examples 

5026 -------- 

5027 >>> import numpy as np 

5028 >>> np.add.nout 

5029 1 

5030 >>> np.multiply.nout 

5031 1 

5032 >>> np.power.nout 

5033 1 

5034 >>> np.exp.nout 

5035 1 

5036 

5037 """)) 

5038 

5039add_newdoc('numpy._core', 'ufunc', ('ntypes', 

5040 """ 

5041 The number of types. 

5042 

5043 The number of numerical NumPy types - of which there are 18 total - on which 

5044 the ufunc can operate. 

5045 

5046 See Also 

5047 -------- 

5048 numpy.ufunc.types 

5049 

5050 Examples 

5051 -------- 

5052 >>> import numpy as np 

5053 >>> np.add.ntypes 

5054 18 

5055 >>> np.multiply.ntypes 

5056 18 

5057 >>> np.power.ntypes 

5058 17 

5059 >>> np.exp.ntypes 

5060 7 

5061 >>> np.remainder.ntypes 

5062 14 

5063 

5064 """)) 

5065 

5066add_newdoc('numpy._core', 'ufunc', ('types', 

5067 """ 

5068 Returns a list with types grouped input->output. 

5069 

5070 Data attribute listing the data-type "Domain-Range" groupings the ufunc can 

5071 deliver. The data-types are given using the character codes. 

5072 

5073 See Also 

5074 -------- 

5075 numpy.ufunc.ntypes 

5076 

5077 Examples 

5078 -------- 

5079 >>> import numpy as np 

5080 >>> np.add.types 

5081 ['??->?', 'bb->b', 'BB->B', 'hh->h', 'HH->H', 'ii->i', 'II->I', 'll->l', 

5082 'LL->L', 'qq->q', 'QQ->Q', 'ff->f', 'dd->d', 'gg->g', 'FF->F', 'DD->D', 

5083 'GG->G', 'OO->O'] 

5084 

5085 >>> np.multiply.types 

5086 ['??->?', 'bb->b', 'BB->B', 'hh->h', 'HH->H', 'ii->i', 'II->I', 'll->l', 

5087 'LL->L', 'qq->q', 'QQ->Q', 'ff->f', 'dd->d', 'gg->g', 'FF->F', 'DD->D', 

5088 'GG->G', 'OO->O'] 

5089 

5090 >>> np.power.types 

5091 ['bb->b', 'BB->B', 'hh->h', 'HH->H', 'ii->i', 'II->I', 'll->l', 'LL->L', 

5092 'qq->q', 'QQ->Q', 'ff->f', 'dd->d', 'gg->g', 'FF->F', 'DD->D', 'GG->G', 

5093 'OO->O'] 

5094 

5095 >>> np.exp.types 

5096 ['f->f', 'd->d', 'g->g', 'F->F', 'D->D', 'G->G', 'O->O'] 

5097 

5098 >>> np.remainder.types 

5099 ['bb->b', 'BB->B', 'hh->h', 'HH->H', 'ii->i', 'II->I', 'll->l', 'LL->L', 

5100 'qq->q', 'QQ->Q', 'ff->f', 'dd->d', 'gg->g', 'OO->O'] 

5101 

5102 """)) 

5103 

5104add_newdoc('numpy._core', 'ufunc', ('signature', 

5105 """ 

5106 Definition of the core elements a generalized ufunc operates on. 

5107 

5108 The signature determines how the dimensions of each input/output array 

5109 are split into core and loop dimensions: 

5110 

5111 1. Each dimension in the signature is matched to a dimension of the 

5112 corresponding passed-in array, starting from the end of the shape tuple. 

5113 2. Core dimensions assigned to the same label in the signature must have 

5114 exactly matching sizes, no broadcasting is performed. 

5115 3. The core dimensions are removed from all inputs and the remaining 

5116 dimensions are broadcast together, defining the loop dimensions. 

5117 

5118 Notes 

5119 ----- 

5120 Generalized ufuncs are used internally in many linalg functions, and in 

5121 the testing suite; the examples below are taken from these. 

5122 For ufuncs that operate on scalars, the signature is None, which is 

5123 equivalent to '()' for every argument. 

5124 

5125 Examples 

5126 -------- 

5127 >>> import numpy as np 

5128 >>> np.linalg._umath_linalg.det.signature 

5129 '(m,m)->()' 

5130 >>> np.matmul.signature 

5131 '(n?,k),(k,m?)->(n?,m?)' 

5132 >>> np.add.signature is None 

5133 True # equivalent to '(),()->()' 

5134 """)) 

5135 

5136############################################################################## 

5137# 

5138# ufunc methods 

5139# 

5140############################################################################## 

5141 

5142add_newdoc('numpy._core', 'ufunc', ('reduce', 

5143 """ 

5144 reduce(array, axis=0, dtype=None, out=None, keepdims=False, initial=<no value>, where=True) 

5145 

5146 Reduces `array`'s dimension by one, by applying ufunc along one axis. 

5147 

5148 Let :math:`array.shape = (N_0, ..., N_i, ..., N_{M-1})`. Then 

5149 :math:`ufunc.reduce(array, axis=i)[k_0, ..,k_{i-1}, k_{i+1}, .., k_{M-1}]` = 

5150 the result of iterating `j` over :math:`range(N_i)`, cumulatively applying 

5151 ufunc to each :math:`array[k_0, ..,k_{i-1}, j, k_{i+1}, .., k_{M-1}]`. 

5152 For a one-dimensional array, reduce produces results equivalent to: 

5153 :: 

5154 

5155 r = op.identity # op = ufunc 

5156 for i in range(len(A)): 

5157 r = op(r, A[i]) 

5158 return r 

5159 

5160 For example, add.reduce() is equivalent to sum(). 

5161 

5162 Parameters 

5163 ---------- 

5164 array : array_like 

5165 The array to act on. 

5166 axis : None or int or tuple of ints, optional 

5167 Axis or axes along which a reduction is performed. 

5168 The default (`axis` = 0) is perform a reduction over the first 

5169 dimension of the input array. `axis` may be negative, in 

5170 which case it counts from the last to the first axis. 

5171 

5172 If this is None, a reduction is performed over all the axes. 

5173 If this is a tuple of ints, a reduction is performed on multiple 

5174 axes, instead of a single axis or all the axes as before. 

5175 

5176 For operations which are either not commutative or not associative, 

5177 doing a reduction over multiple axes is not well-defined. The 

5178 ufuncs do not currently raise an exception in this case, but will 

5179 likely do so in the future. 

5180 dtype : data-type code, optional 

5181 The data type used to perform the operation. Defaults to that of 

5182 ``out`` if given, and the data type of ``array`` otherwise (though 

5183 upcast to conserve precision for some cases, such as 

5184 ``numpy.add.reduce`` for integer or boolean input). 

5185 out : ndarray, None, or tuple of ndarray and None, optional 

5186 A location into which the result is stored. If not provided or None, 

5187 a freshly-allocated array is returned. For consistency with 

5188 ``ufunc.__call__``, if given as a keyword, this may be wrapped in a 

5189 1-element tuple. 

5190 keepdims : bool, optional 

5191 If this is set to True, the axes which are reduced are left 

5192 in the result as dimensions with size one. With this option, 

5193 the result will broadcast correctly against the original `array`. 

5194 initial : scalar, optional 

5195 The value with which to start the reduction. 

5196 If the ufunc has no identity or the dtype is object, this defaults 

5197 to None - otherwise it defaults to ufunc.identity. 

5198 If ``None`` is given, the first element of the reduction is used, 

5199 and an error is thrown if the reduction is empty. 

5200 where : array_like of bool, optional 

5201 A boolean array which is broadcasted to match the dimensions 

5202 of `array`, and selects elements to include in the reduction. Note 

5203 that for ufuncs like ``minimum`` that do not have an identity 

5204 defined, one has to pass in also ``initial``. 

5205 

5206 Returns 

5207 ------- 

5208 r : ndarray 

5209 The reduced array. If `out` was supplied, `r` is a reference to it. 

5210 

5211 Examples 

5212 -------- 

5213 >>> import numpy as np 

5214 >>> np.multiply.reduce([2,3,5]) 

5215 30 

5216 

5217 A multi-dimensional array example: 

5218 

5219 >>> X = np.arange(8).reshape((2,2,2)) 

5220 >>> X 

5221 array([[[0, 1], 

5222 [2, 3]], 

5223 [[4, 5], 

5224 [6, 7]]]) 

5225 >>> np.add.reduce(X, 0) 

5226 array([[ 4, 6], 

5227 [ 8, 10]]) 

5228 >>> np.add.reduce(X) # confirm: default axis value is 0 

5229 array([[ 4, 6], 

5230 [ 8, 10]]) 

5231 >>> np.add.reduce(X, 1) 

5232 array([[ 2, 4], 

5233 [10, 12]]) 

5234 >>> np.add.reduce(X, 2) 

5235 array([[ 1, 5], 

5236 [ 9, 13]]) 

5237 

5238 You can use the ``initial`` keyword argument to initialize the reduction 

5239 with a different value, and ``where`` to select specific elements to include: 

5240 

5241 >>> np.add.reduce([10], initial=5) 

5242 15 

5243 >>> np.add.reduce(np.ones((2, 2, 2)), axis=(0, 2), initial=10) 

5244 array([14., 14.]) 

5245 >>> a = np.array([10., np.nan, 10]) 

5246 >>> np.add.reduce(a, where=~np.isnan(a)) 

5247 20.0 

5248 

5249 Allows reductions of empty arrays where they would normally fail, i.e. 

5250 for ufuncs without an identity. 

5251 

5252 >>> np.minimum.reduce([], initial=np.inf) 

5253 inf 

5254 >>> np.minimum.reduce([[1., 2.], [3., 4.]], initial=10., where=[True, False]) 

5255 array([ 1., 10.]) 

5256 >>> np.minimum.reduce([]) 

5257 Traceback (most recent call last): 

5258 ... 

5259 ValueError: zero-size array to reduction operation minimum which has no identity 

5260 """)) 

5261 

5262add_newdoc('numpy._core', 'ufunc', ('accumulate', 

5263 """ 

5264 accumulate(array, axis=0, dtype=None, out=None) 

5265 

5266 Accumulate the result of applying the operator to all elements. 

5267 

5268 For a one-dimensional array, accumulate produces results equivalent to:: 

5269 

5270 r = np.empty(len(A)) 

5271 t = op.identity # op = the ufunc being applied to A's elements 

5272 for i in range(len(A)): 

5273 t = op(t, A[i]) 

5274 r[i] = t 

5275 return r 

5276 

5277 For example, add.accumulate() is equivalent to np.cumsum(). 

5278 

5279 For a multi-dimensional array, accumulate is applied along only one 

5280 axis (axis zero by default; see Examples below) so repeated use is 

5281 necessary if one wants to accumulate over multiple axes. 

5282 

5283 Parameters 

5284 ---------- 

5285 array : array_like 

5286 The array to act on. 

5287 axis : int, optional 

5288 The axis along which to apply the accumulation; default is zero. 

5289 dtype : data-type code, optional 

5290 The data-type used to represent the intermediate results. Defaults 

5291 to the data-type of the output array if such is provided, or the 

5292 data-type of the input array if no output array is provided. 

5293 out : ndarray, None, or tuple of ndarray and None, optional 

5294 A location into which the result is stored. If not provided or None, 

5295 a freshly-allocated array is returned. For consistency with 

5296 ``ufunc.__call__``, if given as a keyword, this may be wrapped in a 

5297 1-element tuple. 

5298 

5299 Returns 

5300 ------- 

5301 r : ndarray 

5302 The accumulated values. If `out` was supplied, `r` is a reference to 

5303 `out`. 

5304 

5305 Examples 

5306 -------- 

5307 1-D array examples: 

5308 

5309 >>> import numpy as np 

5310 >>> np.add.accumulate([2, 3, 5]) 

5311 array([ 2, 5, 10]) 

5312 >>> np.multiply.accumulate([2, 3, 5]) 

5313 array([ 2, 6, 30]) 

5314 

5315 2-D array examples: 

5316 

5317 >>> I = np.eye(2) 

5318 >>> I 

5319 array([[1., 0.], 

5320 [0., 1.]]) 

5321 

5322 Accumulate along axis 0 (rows), down columns: 

5323 

5324 >>> np.add.accumulate(I, 0) 

5325 array([[1., 0.], 

5326 [1., 1.]]) 

5327 >>> np.add.accumulate(I) # no axis specified = axis zero 

5328 array([[1., 0.], 

5329 [1., 1.]]) 

5330 

5331 Accumulate along axis 1 (columns), through rows: 

5332 

5333 >>> np.add.accumulate(I, 1) 

5334 array([[1., 1.], 

5335 [0., 1.]]) 

5336 

5337 """)) 

5338 

5339add_newdoc('numpy._core', 'ufunc', ('reduceat', 

5340 """ 

5341 reduceat(array, indices, axis=0, dtype=None, out=None) 

5342 

5343 Performs a (local) reduce with specified slices over a single axis. 

5344 

5345 For i in ``range(len(indices))``, `reduceat` computes 

5346 ``ufunc.reduce(array[indices[i]:indices[i+1]])``, which becomes the i-th 

5347 generalized "row" parallel to `axis` in the final result (i.e., in a 

5348 2-D array, for example, if `axis = 0`, it becomes the i-th row, but if 

5349 `axis = 1`, it becomes the i-th column). There are three exceptions to this: 

5350 

5351 * when ``i = len(indices) - 1`` (so for the last index), 

5352 ``indices[i+1] = array.shape[axis]``. 

5353 * if ``indices[i] >= indices[i + 1]``, the i-th generalized "row" is 

5354 simply ``array[indices[i]]``. 

5355 * if ``indices[i] >= len(array)`` or ``indices[i] < 0``, an error is raised. 

5356 

5357 The shape of the output depends on the size of `indices`, and may be 

5358 larger than `array` (this happens if ``len(indices) > array.shape[axis]``). 

5359 

5360 Parameters 

5361 ---------- 

5362 array : array_like 

5363 The array to act on. 

5364 indices : array_like 

5365 Paired indices, comma separated (not colon), specifying slices to 

5366 reduce. 

5367 axis : int, optional 

5368 The axis along which to apply the reduceat. 

5369 dtype : data-type code, optional 

5370 The data type used to perform the operation. Defaults to that of 

5371 ``out`` if given, and the data type of ``array`` otherwise (though 

5372 upcast to conserve precision for some cases, such as 

5373 ``numpy.add.reduce`` for integer or boolean input). 

5374 out : ndarray, None, or tuple of ndarray and None, optional 

5375 A location into which the result is stored. If not provided or None, 

5376 a freshly-allocated array is returned. For consistency with 

5377 ``ufunc.__call__``, if given as a keyword, this may be wrapped in a 

5378 1-element tuple. 

5379 

5380 Returns 

5381 ------- 

5382 r : ndarray 

5383 The reduced values. If `out` was supplied, `r` is a reference to 

5384 `out`. 

5385 

5386 Notes 

5387 ----- 

5388 A descriptive example: 

5389 

5390 If `array` is 1-D, the function `ufunc.accumulate(array)` is the same as 

5391 ``ufunc.reduceat(array, indices)[::2]`` where `indices` is 

5392 ``range(len(array) - 1)`` with a zero placed 

5393 in every other element: 

5394 ``indices = zeros(2 * len(array) - 1)``, 

5395 ``indices[1::2] = range(1, len(array))``. 

5396 

5397 Don't be fooled by this attribute's name: `reduceat(array)` is not 

5398 necessarily smaller than `array`. 

5399 

5400 Examples 

5401 -------- 

5402 To take the running sum of four successive values: 

5403 

5404 >>> import numpy as np 

5405 >>> np.add.reduceat(np.arange(8),[0,4, 1,5, 2,6, 3,7])[::2] 

5406 array([ 6, 10, 14, 18]) 

5407 

5408 A 2-D example: 

5409 

5410 >>> x = np.linspace(0, 15, 16).reshape(4,4) 

5411 >>> x 

5412 array([[ 0., 1., 2., 3.], 

5413 [ 4., 5., 6., 7.], 

5414 [ 8., 9., 10., 11.], 

5415 [12., 13., 14., 15.]]) 

5416 

5417 :: 

5418 

5419 # reduce such that the result has the following five rows: 

5420 # [row1 + row2 + row3] 

5421 # [row4] 

5422 # [row2] 

5423 # [row3] 

5424 # [row1 + row2 + row3 + row4] 

5425 

5426 >>> np.add.reduceat(x, [0, 3, 1, 2, 0]) 

5427 array([[12., 15., 18., 21.], 

5428 [12., 13., 14., 15.], 

5429 [ 4., 5., 6., 7.], 

5430 [ 8., 9., 10., 11.], 

5431 [24., 28., 32., 36.]]) 

5432 

5433 :: 

5434 

5435 # reduce such that result has the following two columns: 

5436 # [col1 * col2 * col3, col4] 

5437 

5438 >>> np.multiply.reduceat(x, [0, 3], 1) 

5439 array([[ 0., 3.], 

5440 [ 120., 7.], 

5441 [ 720., 11.], 

5442 [2184., 15.]]) 

5443 

5444 """)) 

5445 

5446add_newdoc('numpy._core', 'ufunc', ('outer', 

5447 r""" 

5448 outer(A, B, /, **kwargs) 

5449 

5450 Apply the ufunc `op` to all pairs (a, b) with a in `A` and b in `B`. 

5451 

5452 Let ``M = A.ndim``, ``N = B.ndim``. Then the result, `C`, of 

5453 ``op.outer(A, B)`` is an array of dimension M + N such that: 

5454 

5455 .. math:: C[i_0, ..., i_{M-1}, j_0, ..., j_{N-1}] = 

5456 op(A[i_0, ..., i_{M-1}], B[j_0, ..., j_{N-1}]) 

5457 

5458 For `A` and `B` one-dimensional, this is equivalent to:: 

5459 

5460 r = empty(len(A),len(B)) 

5461 for i in range(len(A)): 

5462 for j in range(len(B)): 

5463 r[i,j] = op(A[i], B[j]) # op = ufunc in question 

5464 

5465 Parameters 

5466 ---------- 

5467 A : array_like 

5468 First array 

5469 B : array_like 

5470 Second array 

5471 kwargs : any 

5472 Arguments to pass on to the ufunc. Typically `dtype` or `out`. 

5473 See `ufunc` for a comprehensive overview of all available arguments. 

5474 

5475 Returns 

5476 ------- 

5477 r : ndarray 

5478 Output array 

5479 

5480 See Also 

5481 -------- 

5482 numpy.outer : A less powerful version of ``np.multiply.outer`` 

5483 that `ravel`\ s all inputs to 1D. This exists 

5484 primarily for compatibility with old code. 

5485 

5486 tensordot : ``np.tensordot(a, b, axes=((), ()))`` and 

5487 ``np.multiply.outer(a, b)`` behave same for all 

5488 dimensions of a and b. 

5489 

5490 Examples 

5491 -------- 

5492 >>> np.multiply.outer([1, 2, 3], [4, 5, 6]) 

5493 array([[ 4, 5, 6], 

5494 [ 8, 10, 12], 

5495 [12, 15, 18]]) 

5496 

5497 A multi-dimensional example: 

5498 

5499 >>> A = np.array([[1, 2, 3], [4, 5, 6]]) 

5500 >>> A.shape 

5501 (2, 3) 

5502 >>> B = np.array([[1, 2, 3, 4]]) 

5503 >>> B.shape 

5504 (1, 4) 

5505 >>> C = np.multiply.outer(A, B) 

5506 >>> C.shape; C 

5507 (2, 3, 1, 4) 

5508 array([[[[ 1, 2, 3, 4]], 

5509 [[ 2, 4, 6, 8]], 

5510 [[ 3, 6, 9, 12]]], 

5511 [[[ 4, 8, 12, 16]], 

5512 [[ 5, 10, 15, 20]], 

5513 [[ 6, 12, 18, 24]]]]) 

5514 

5515 """)) 

5516 

5517add_newdoc('numpy._core', 'ufunc', ('at', 

5518 """ 

5519 at(a, indices, b=None, /) 

5520 

5521 Performs unbuffered in place operation on operand 'a' for elements 

5522 specified by 'indices'. For addition ufunc, this method is equivalent to 

5523 ``a[indices] += b``, except that results are accumulated for elements that 

5524 are indexed more than once. For example, ``a[[0,0]] += 1`` will only 

5525 increment the first element once because of buffering, whereas 

5526 ``add.at(a, [0,0], 1)`` will increment the first element twice. 

5527 

5528 Parameters 

5529 ---------- 

5530 a : array_like 

5531 The array to perform in place operation on. 

5532 indices : array_like or tuple 

5533 Array like index object or slice object for indexing into first 

5534 operand. If first operand has multiple dimensions, indices can be a 

5535 tuple of array like index objects or slice objects. 

5536 b : array_like 

5537 Second operand for ufuncs requiring two operands. Operand must be 

5538 broadcastable over first operand after indexing or slicing. 

5539 

5540 Examples 

5541 -------- 

5542 Set items 0 and 1 to their negative values: 

5543 

5544 >>> import numpy as np 

5545 >>> a = np.array([1, 2, 3, 4]) 

5546 >>> np.negative.at(a, [0, 1]) 

5547 >>> a 

5548 array([-1, -2, 3, 4]) 

5549 

5550 Increment items 0 and 1, and increment item 2 twice: 

5551 

5552 >>> a = np.array([1, 2, 3, 4]) 

5553 >>> np.add.at(a, [0, 1, 2, 2], 1) 

5554 >>> a 

5555 array([2, 3, 5, 4]) 

5556 

5557 Add items 0 and 1 in first array to second array, 

5558 and store results in first array: 

5559 

5560 >>> a = np.array([1, 2, 3, 4]) 

5561 >>> b = np.array([1, 2]) 

5562 >>> np.add.at(a, [0, 1], b) 

5563 >>> a 

5564 array([2, 4, 3, 4]) 

5565 

5566 """)) 

5567 

5568add_newdoc('numpy._core', 'ufunc', ('resolve_dtypes', 

5569 """ 

5570 resolve_dtypes(dtypes, *, signature=None, casting=None, reduction=False) 

5571 

5572 Find the dtypes NumPy will use for the operation. Both input and 

5573 output dtypes are returned and may differ from those provided. 

5574 

5575 .. note:: 

5576 

5577 This function always applies NEP 50 rules since it is not provided 

5578 any actual values. The Python types ``int``, ``float``, and 

5579 ``complex`` thus behave weak and should be passed for "untyped" 

5580 Python input. 

5581 

5582 Parameters 

5583 ---------- 

5584 dtypes : tuple of dtypes, None, or literal int, float, complex 

5585 The input dtypes for each operand. Output operands can be 

5586 None, indicating that the dtype must be found. 

5587 signature : tuple of DTypes or None, optional 

5588 If given, enforces exact DType (classes) of the specific operand. 

5589 The ufunc ``dtype`` argument is equivalent to passing a tuple with 

5590 only output dtypes set. 

5591 casting : {'no', 'equiv', 'safe', 'same_kind', 'unsafe'}, optional 

5592 The casting mode when casting is necessary. This is identical to 

5593 the ufunc call casting modes. 

5594 reduction : boolean 

5595 If given, the resolution assumes a reduce operation is happening 

5596 which slightly changes the promotion and type resolution rules. 

5597 `dtypes` is usually something like ``(None, np.dtype("i2"), None)`` 

5598 for reductions (first input is also the output). 

5599 

5600 .. note:: 

5601 

5602 The default casting mode is "same_kind", however, as of 

5603 NumPy 1.24, NumPy uses "unsafe" for reductions. 

5604 

5605 Returns 

5606 ------- 

5607 dtypes : tuple of dtypes 

5608 The dtypes which NumPy would use for the calculation. Note that 

5609 dtypes may not match the passed in ones (casting is necessary). 

5610 

5611 

5612 Examples 

5613 -------- 

5614 This API requires passing dtypes, define them for convenience: 

5615 

5616 >>> import numpy as np 

5617 >>> int32 = np.dtype("int32") 

5618 >>> float32 = np.dtype("float32") 

5619 

5620 The typical ufunc call does not pass an output dtype. `numpy.add` has two 

5621 inputs and one output, so leave the output as ``None`` (not provided): 

5622 

5623 >>> np.add.resolve_dtypes((int32, float32, None)) 

5624 (dtype('float64'), dtype('float64'), dtype('float64')) 

5625 

5626 The loop found uses "float64" for all operands (including the output), the 

5627 first input would be cast. 

5628 

5629 ``resolve_dtypes`` supports "weak" handling for Python scalars by passing 

5630 ``int``, ``float``, or ``complex``: 

5631 

5632 >>> np.add.resolve_dtypes((float32, float, None)) 

5633 (dtype('float32'), dtype('float32'), dtype('float32')) 

5634 

5635 Where the Python ``float`` behaves similar to a Python value ``0.0`` 

5636 in a ufunc call. (See :ref:`NEP 50 <NEP50>` for details.) 

5637 

5638 """)) 

5639 

5640add_newdoc('numpy._core', 'ufunc', ('_resolve_dtypes_and_context', 

5641 """ 

5642 _resolve_dtypes_and_context(dtypes, *, signature=None, casting=None, reduction=False) 

5643 

5644 See `numpy.ufunc.resolve_dtypes` for parameter information. This 

5645 function is considered *unstable*. You may use it, but the returned 

5646 information is NumPy version specific and expected to change. 

5647 Large API/ABI changes are not expected, but a new NumPy version is 

5648 expected to require updating code using this functionality. 

5649 

5650 This function is designed to be used in conjunction with 

5651 `numpy.ufunc._get_strided_loop`. The calls are split to mirror the C API 

5652 and allow future improvements. 

5653 

5654 Returns 

5655 ------- 

5656 dtypes : tuple of dtypes 

5657 call_info : 

5658 PyCapsule with all necessary information to get access to low level 

5659 C calls. See `numpy.ufunc._get_strided_loop` for more information. 

5660 

5661 """)) 

5662 

5663add_newdoc('numpy._core', 'ufunc', ('_get_strided_loop', 

5664 """ 

5665 _get_strided_loop(call_info, /, *, fixed_strides=None) 

5666 

5667 This function fills in the ``call_info`` capsule to include all 

5668 information necessary to call the low-level strided loop from NumPy. 

5669 

5670 See notes for more information. 

5671 

5672 Parameters 

5673 ---------- 

5674 call_info : PyCapsule 

5675 The PyCapsule returned by `numpy.ufunc._resolve_dtypes_and_context`. 

5676 fixed_strides : tuple of int or None, optional 

5677 A tuple with fixed byte strides of all input arrays. NumPy may use 

5678 this information to find specialized loops, so any call must follow 

5679 the given stride. Use ``None`` to indicate that the stride is not 

5680 known (or not fixed) for all calls. 

5681 

5682 Notes 

5683 ----- 

5684 Together with `numpy.ufunc._resolve_dtypes_and_context` this function 

5685 gives low-level access to the NumPy ufunc loops. 

5686 The first function does general preparation and returns the required 

5687 information. It returns this as a C capsule with the version specific 

5688 name ``numpy_1.24_ufunc_call_info``. 

5689 The NumPy 1.24 ufunc call info capsule has the following layout:: 

5690 

5691 typedef struct { 

5692 PyArrayMethod_StridedLoop *strided_loop; 

5693 PyArrayMethod_Context *context; 

5694 NpyAuxData *auxdata; 

5695 

5696 /* Flag information (expected to change) */ 

5697 npy_bool requires_pyapi; /* GIL is required by loop */ 

5698 

5699 /* Loop doesn't set FPE flags; if not set check FPE flags */ 

5700 npy_bool no_floatingpoint_errors; 

5701 } ufunc_call_info; 

5702 

5703 Note that the first call only fills in the ``context``. The call to 

5704 ``_get_strided_loop`` fills in all other data. The main thing to note is 

5705 that the new-style loops return 0 on success, -1 on failure. They are 

5706 passed context as new first input and ``auxdata`` as (replaced) last. 

5707 

5708 Only the ``strided_loop``signature is considered guaranteed stable 

5709 for NumPy bug-fix releases. All other API is tied to the experimental 

5710 API versioning. 

5711 

5712 The reason for the split call is that cast information is required to 

5713 decide what the fixed-strides will be. 

5714 

5715 NumPy ties the lifetime of the ``auxdata`` information to the capsule. 

5716 

5717 """)) 

5718 

5719 

5720 

5721############################################################################## 

5722# 

5723# Documentation for dtype attributes and methods 

5724# 

5725############################################################################## 

5726 

5727############################################################################## 

5728# 

5729# dtype object 

5730# 

5731############################################################################## 

5732 

5733add_newdoc('numpy._core.multiarray', 'dtype', 

5734 """ 

5735 dtype(dtype, align=False, copy=False, [metadata]) 

5736 

5737 Create a data type object. 

5738 

5739 A numpy array is homogeneous, and contains elements described by a 

5740 dtype object. A dtype object can be constructed from different 

5741 combinations of fundamental numeric types. 

5742 

5743 Parameters 

5744 ---------- 

5745 dtype 

5746 Object to be converted to a data type object. 

5747 align : bool, optional 

5748 Add padding to the fields to match what a C compiler would output 

5749 for a similar C-struct. Can be ``True`` only if `obj` is a dictionary 

5750 or a comma-separated string. If a struct dtype is being created, 

5751 this also sets a sticky alignment flag ``isalignedstruct``. 

5752 copy : bool, optional 

5753 Make a new copy of the data-type object. If ``False``, the result 

5754 may just be a reference to a built-in data-type object. 

5755 metadata : dict, optional 

5756 An optional dictionary with dtype metadata. 

5757 

5758 See also 

5759 -------- 

5760 result_type 

5761 

5762 Examples 

5763 -------- 

5764 Using array-scalar type: 

5765 

5766 >>> import numpy as np 

5767 >>> np.dtype(np.int16) 

5768 dtype('int16') 

5769 

5770 Structured type, one field name 'f1', containing int16: 

5771 

5772 >>> np.dtype([('f1', np.int16)]) 

5773 dtype([('f1', '<i2')]) 

5774 

5775 Structured type, one field named 'f1', in itself containing a structured 

5776 type with one field: 

5777 

5778 >>> np.dtype([('f1', [('f1', np.int16)])]) 

5779 dtype([('f1', [('f1', '<i2')])]) 

5780 

5781 Structured type, two fields: the first field contains an unsigned int, the 

5782 second an int32: 

5783 

5784 >>> np.dtype([('f1', np.uint64), ('f2', np.int32)]) 

5785 dtype([('f1', '<u8'), ('f2', '<i4')]) 

5786 

5787 Using array-protocol type strings: 

5788 

5789 >>> np.dtype([('a','f8'),('b','S10')]) 

5790 dtype([('a', '<f8'), ('b', 'S10')]) 

5791 

5792 Using comma-separated field formats. The shape is (2,3): 

5793 

5794 >>> np.dtype("i4, (2,3)f8") 

5795 dtype([('f0', '<i4'), ('f1', '<f8', (2, 3))]) 

5796 

5797 Using tuples. ``int`` is a fixed type, 3 the field's shape. ``void`` 

5798 is a flexible type, here of size 10: 

5799 

5800 >>> np.dtype([('hello',(np.int64,3)),('world',np.void,10)]) 

5801 dtype([('hello', '<i8', (3,)), ('world', 'V10')]) 

5802 

5803 Subdivide ``int16`` into 2 ``int8``'s, called x and y. 0 and 1 are 

5804 the offsets in bytes: 

5805 

5806 >>> np.dtype((np.int16, {'x':(np.int8,0), 'y':(np.int8,1)})) 

5807 dtype((numpy.int16, [('x', 'i1'), ('y', 'i1')])) 

5808 

5809 Using dictionaries. Two fields named 'gender' and 'age': 

5810 

5811 >>> np.dtype({'names':['gender','age'], 'formats':['S1',np.uint8]}) 

5812 dtype([('gender', 'S1'), ('age', 'u1')]) 

5813 

5814 Offsets in bytes, here 0 and 25: 

5815 

5816 >>> np.dtype({'surname':('S25',0),'age':(np.uint8,25)}) 

5817 dtype([('surname', 'S25'), ('age', 'u1')]) 

5818 

5819 """) 

5820 

5821############################################################################## 

5822# 

5823# dtype attributes 

5824# 

5825############################################################################## 

5826 

5827add_newdoc('numpy._core.multiarray', 'dtype', ('alignment', 

5828 """ 

5829 The required alignment (bytes) of this data-type according to the compiler. 

5830 

5831 More information is available in the C-API section of the manual. 

5832 

5833 Examples 

5834 -------- 

5835 

5836 >>> import numpy as np 

5837 >>> x = np.dtype('i4') 

5838 >>> x.alignment 

5839 4 

5840 

5841 >>> x = np.dtype(float) 

5842 >>> x.alignment 

5843 8 

5844 

5845 """)) 

5846 

5847add_newdoc('numpy._core.multiarray', 'dtype', ('byteorder', 

5848 """ 

5849 A character indicating the byte-order of this data-type object. 

5850 

5851 One of: 

5852 

5853 === ============== 

5854 '=' native 

5855 '<' little-endian 

5856 '>' big-endian 

5857 '|' not applicable 

5858 === ============== 

5859 

5860 All built-in data-type objects have byteorder either '=' or '|'. 

5861 

5862 Examples 

5863 -------- 

5864 

5865 >>> import numpy as np 

5866 >>> dt = np.dtype('i2') 

5867 >>> dt.byteorder 

5868 '=' 

5869 >>> # endian is not relevant for 8 bit numbers 

5870 >>> np.dtype('i1').byteorder 

5871 '|' 

5872 >>> # or ASCII strings 

5873 >>> np.dtype('S2').byteorder 

5874 '|' 

5875 >>> # Even if specific code is given, and it is native 

5876 >>> # '=' is the byteorder 

5877 >>> import sys 

5878 >>> sys_is_le = sys.byteorder == 'little' 

5879 >>> native_code = '<' if sys_is_le else '>' 

5880 >>> swapped_code = '>' if sys_is_le else '<' 

5881 >>> dt = np.dtype(native_code + 'i2') 

5882 >>> dt.byteorder 

5883 '=' 

5884 >>> # Swapped code shows up as itself 

5885 >>> dt = np.dtype(swapped_code + 'i2') 

5886 >>> dt.byteorder == swapped_code 

5887 True 

5888 

5889 """)) 

5890 

5891add_newdoc('numpy._core.multiarray', 'dtype', ('char', 

5892 """A unique character code for each of the 21 different built-in types. 

5893 

5894 Examples 

5895 -------- 

5896 

5897 >>> import numpy as np 

5898 >>> x = np.dtype(float) 

5899 >>> x.char 

5900 'd' 

5901 

5902 """)) 

5903 

5904add_newdoc('numpy._core.multiarray', 'dtype', ('descr', 

5905 """ 

5906 `__array_interface__` description of the data-type. 

5907 

5908 The format is that required by the 'descr' key in the 

5909 `__array_interface__` attribute. 

5910 

5911 Warning: This attribute exists specifically for `__array_interface__`, 

5912 and passing it directly to `numpy.dtype` will not accurately reconstruct 

5913 some dtypes (e.g., scalar and subarray dtypes). 

5914 

5915 Examples 

5916 -------- 

5917 

5918 >>> import numpy as np 

5919 >>> x = np.dtype(float) 

5920 >>> x.descr 

5921 [('', '<f8')] 

5922 

5923 >>> dt = np.dtype([('name', np.str_, 16), ('grades', np.float64, (2,))]) 

5924 >>> dt.descr 

5925 [('name', '<U16'), ('grades', '<f8', (2,))] 

5926 

5927 """)) 

5928 

5929add_newdoc('numpy._core.multiarray', 'dtype', ('fields', 

5930 """ 

5931 Dictionary of named fields defined for this data type, or ``None``. 

5932 

5933 The dictionary is indexed by keys that are the names of the fields. 

5934 Each entry in the dictionary is a tuple fully describing the field:: 

5935 

5936 (dtype, offset[, title]) 

5937 

5938 Offset is limited to C int, which is signed and usually 32 bits. 

5939 If present, the optional title can be any object (if it is a string 

5940 or unicode then it will also be a key in the fields dictionary, 

5941 otherwise it's meta-data). Notice also that the first two elements 

5942 of the tuple can be passed directly as arguments to the 

5943 ``ndarray.getfield`` and ``ndarray.setfield`` methods. 

5944 

5945 See Also 

5946 -------- 

5947 ndarray.getfield, ndarray.setfield 

5948 

5949 Examples 

5950 -------- 

5951 

5952 >>> import numpy as np 

5953 >>> dt = np.dtype([('name', np.str_, 16), ('grades', np.float64, (2,))]) 

5954 >>> print(dt.fields) 

5955 {'grades': (dtype(('float64',(2,))), 16), 'name': (dtype('|S16'), 0)} 

5956 

5957 """)) 

5958 

5959add_newdoc('numpy._core.multiarray', 'dtype', ('flags', 

5960 """ 

5961 Bit-flags describing how this data type is to be interpreted. 

5962 

5963 Bit-masks are in ``numpy._core.multiarray`` as the constants 

5964 `ITEM_HASOBJECT`, `LIST_PICKLE`, `ITEM_IS_POINTER`, `NEEDS_INIT`, 

5965 `NEEDS_PYAPI`, `USE_GETITEM`, `USE_SETITEM`. A full explanation 

5966 of these flags is in C-API documentation; they are largely useful 

5967 for user-defined data-types. 

5968 

5969 The following example demonstrates that operations on this particular 

5970 dtype requires Python C-API. 

5971 

5972 Examples 

5973 -------- 

5974 

5975 >>> import numpy as np 

5976 >>> x = np.dtype([('a', np.int32, 8), ('b', np.float64, 6)]) 

5977 >>> x.flags 

5978 16 

5979 >>> np._core.multiarray.NEEDS_PYAPI 

5980 16 

5981 

5982 """)) 

5983 

5984add_newdoc('numpy._core.multiarray', 'dtype', ('hasobject', 

5985 """ 

5986 Boolean indicating whether this dtype contains any reference-counted 

5987 objects in any fields or sub-dtypes. 

5988 

5989 Recall that what is actually in the ndarray memory representing 

5990 the Python object is the memory address of that object (a pointer). 

5991 Special handling may be required, and this attribute is useful for 

5992 distinguishing data types that may contain arbitrary Python objects 

5993 and data-types that won't. 

5994 

5995 """)) 

5996 

5997add_newdoc('numpy._core.multiarray', 'dtype', ('isbuiltin', 

5998 """ 

5999 Integer indicating how this dtype relates to the built-in dtypes. 

6000 

6001 Read-only. 

6002 

6003 = ======================================================================== 

6004 0 if this is a structured array type, with fields 

6005 1 if this is a dtype compiled into numpy (such as ints, floats etc) 

6006 2 if the dtype is for a user-defined numpy type 

6007 A user-defined type uses the numpy C-API machinery to extend 

6008 numpy to handle a new array type. See 

6009 :ref:`user.user-defined-data-types` in the NumPy manual. 

6010 = ======================================================================== 

6011 

6012 Examples 

6013 -------- 

6014 

6015 >>> import numpy as np 

6016 >>> dt = np.dtype('i2') 

6017 >>> dt.isbuiltin 

6018 1 

6019 >>> dt = np.dtype('f8') 

6020 >>> dt.isbuiltin 

6021 1 

6022 >>> dt = np.dtype([('field1', 'f8')]) 

6023 >>> dt.isbuiltin 

6024 0 

6025 

6026 """)) 

6027 

6028add_newdoc('numpy._core.multiarray', 'dtype', ('isnative', 

6029 """ 

6030 Boolean indicating whether the byte order of this dtype is native 

6031 to the platform. 

6032 

6033 """)) 

6034 

6035add_newdoc('numpy._core.multiarray', 'dtype', ('isalignedstruct', 

6036 """ 

6037 Boolean indicating whether the dtype is a struct which maintains 

6038 field alignment. This flag is sticky, so when combining multiple 

6039 structs together, it is preserved and produces new dtypes which 

6040 are also aligned. 

6041 

6042 """)) 

6043 

6044add_newdoc('numpy._core.multiarray', 'dtype', ('itemsize', 

6045 """ 

6046 The element size of this data-type object. 

6047 

6048 For 18 of the 21 types this number is fixed by the data-type. 

6049 For the flexible data-types, this number can be anything. 

6050 

6051 Examples 

6052 -------- 

6053 

6054 >>> import numpy as np 

6055 >>> arr = np.array([[1, 2], [3, 4]]) 

6056 >>> arr.dtype 

6057 dtype('int64') 

6058 >>> arr.itemsize 

6059 8 

6060 

6061 >>> dt = np.dtype([('name', np.str_, 16), ('grades', np.float64, (2,))]) 

6062 >>> dt.itemsize 

6063 80 

6064 

6065 """)) 

6066 

6067add_newdoc('numpy._core.multiarray', 'dtype', ('kind', 

6068 """ 

6069 A character code (one of 'biufcmMOSUV') identifying the general kind of data. 

6070 

6071 = ====================== 

6072 b boolean 

6073 i signed integer 

6074 u unsigned integer 

6075 f floating-point 

6076 c complex floating-point 

6077 m timedelta 

6078 M datetime 

6079 O object 

6080 S (byte-)string 

6081 U Unicode 

6082 V void 

6083 = ====================== 

6084 

6085 Examples 

6086 -------- 

6087 

6088 >>> import numpy as np 

6089 >>> dt = np.dtype('i4') 

6090 >>> dt.kind 

6091 'i' 

6092 >>> dt = np.dtype('f8') 

6093 >>> dt.kind 

6094 'f' 

6095 >>> dt = np.dtype([('field1', 'f8')]) 

6096 >>> dt.kind 

6097 'V' 

6098 

6099 """)) 

6100 

6101add_newdoc('numpy._core.multiarray', 'dtype', ('metadata', 

6102 """ 

6103 Either ``None`` or a readonly dictionary of metadata (mappingproxy). 

6104 

6105 The metadata field can be set using any dictionary at data-type 

6106 creation. NumPy currently has no uniform approach to propagating 

6107 metadata; although some array operations preserve it, there is no 

6108 guarantee that others will. 

6109 

6110 .. warning:: 

6111 

6112 Although used in certain projects, this feature was long undocumented 

6113 and is not well supported. Some aspects of metadata propagation 

6114 are expected to change in the future. 

6115 

6116 Examples 

6117 -------- 

6118 

6119 >>> import numpy as np 

6120 >>> dt = np.dtype(float, metadata={"key": "value"}) 

6121 >>> dt.metadata["key"] 

6122 'value' 

6123 >>> arr = np.array([1, 2, 3], dtype=dt) 

6124 >>> arr.dtype.metadata 

6125 mappingproxy({'key': 'value'}) 

6126 

6127 Adding arrays with identical datatypes currently preserves the metadata: 

6128 

6129 >>> (arr + arr).dtype.metadata 

6130 mappingproxy({'key': 'value'}) 

6131 

6132 But if the arrays have different dtype metadata, the metadata may be 

6133 dropped: 

6134 

6135 >>> dt2 = np.dtype(float, metadata={"key2": "value2"}) 

6136 >>> arr2 = np.array([3, 2, 1], dtype=dt2) 

6137 >>> (arr + arr2).dtype.metadata is None 

6138 True # The metadata field is cleared so None is returned 

6139 """)) 

6140 

6141add_newdoc('numpy._core.multiarray', 'dtype', ('name', 

6142 """ 

6143 A bit-width name for this data-type. 

6144 

6145 Un-sized flexible data-type objects do not have this attribute. 

6146 

6147 Examples 

6148 -------- 

6149 

6150 >>> import numpy as np 

6151 >>> x = np.dtype(float) 

6152 >>> x.name 

6153 'float64' 

6154 >>> x = np.dtype([('a', np.int32, 8), ('b', np.float64, 6)]) 

6155 >>> x.name 

6156 'void640' 

6157 

6158 """)) 

6159 

6160add_newdoc('numpy._core.multiarray', 'dtype', ('names', 

6161 """ 

6162 Ordered list of field names, or ``None`` if there are no fields. 

6163 

6164 The names are ordered according to increasing byte offset. This can be 

6165 used, for example, to walk through all of the named fields in offset order. 

6166 

6167 Examples 

6168 -------- 

6169 >>> dt = np.dtype([('name', np.str_, 16), ('grades', np.float64, (2,))]) 

6170 >>> dt.names 

6171 ('name', 'grades') 

6172 

6173 """)) 

6174 

6175add_newdoc('numpy._core.multiarray', 'dtype', ('num', 

6176 """ 

6177 A unique number for each of the 21 different built-in types. 

6178 

6179 These are roughly ordered from least-to-most precision. 

6180 

6181 Examples 

6182 -------- 

6183 

6184 >>> import numpy as np 

6185 >>> dt = np.dtype(str) 

6186 >>> dt.num 

6187 19 

6188 

6189 >>> dt = np.dtype(float) 

6190 >>> dt.num 

6191 12 

6192 

6193 """)) 

6194 

6195add_newdoc('numpy._core.multiarray', 'dtype', ('shape', 

6196 """ 

6197 Shape tuple of the sub-array if this data type describes a sub-array, 

6198 and ``()`` otherwise. 

6199 

6200 Examples 

6201 -------- 

6202 

6203 >>> import numpy as np 

6204 >>> dt = np.dtype(('i4', 4)) 

6205 >>> dt.shape 

6206 (4,) 

6207 

6208 >>> dt = np.dtype(('i4', (2, 3))) 

6209 >>> dt.shape 

6210 (2, 3) 

6211 

6212 """)) 

6213 

6214add_newdoc('numpy._core.multiarray', 'dtype', ('ndim', 

6215 """ 

6216 Number of dimensions of the sub-array if this data type describes a 

6217 sub-array, and ``0`` otherwise. 

6218 

6219 Examples 

6220 -------- 

6221 >>> import numpy as np 

6222 >>> x = np.dtype(float) 

6223 >>> x.ndim 

6224 0 

6225 

6226 >>> x = np.dtype((float, 8)) 

6227 >>> x.ndim 

6228 1 

6229 

6230 >>> x = np.dtype(('i4', (3, 4))) 

6231 >>> x.ndim 

6232 2 

6233 

6234 """)) 

6235 

6236add_newdoc('numpy._core.multiarray', 'dtype', ('str', 

6237 """The array-protocol typestring of this data-type object.""")) 

6238 

6239add_newdoc('numpy._core.multiarray', 'dtype', ('subdtype', 

6240 """ 

6241 Tuple ``(item_dtype, shape)`` if this `dtype` describes a sub-array, and 

6242 None otherwise. 

6243 

6244 The *shape* is the fixed shape of the sub-array described by this 

6245 data type, and *item_dtype* the data type of the array. 

6246 

6247 If a field whose dtype object has this attribute is retrieved, 

6248 then the extra dimensions implied by *shape* are tacked on to 

6249 the end of the retrieved array. 

6250 

6251 See Also 

6252 -------- 

6253 dtype.base 

6254 

6255 Examples 

6256 -------- 

6257 >>> import numpy as np 

6258 >>> x = numpy.dtype('8f') 

6259 >>> x.subdtype 

6260 (dtype('float32'), (8,)) 

6261 

6262 >>> x = numpy.dtype('i2') 

6263 >>> x.subdtype 

6264 >>> 

6265 

6266 """)) 

6267 

6268add_newdoc('numpy._core.multiarray', 'dtype', ('base', 

6269 """ 

6270 Returns dtype for the base element of the subarrays, 

6271 regardless of their dimension or shape. 

6272 

6273 See Also 

6274 -------- 

6275 dtype.subdtype 

6276 

6277 Examples 

6278 -------- 

6279 >>> import numpy as np 

6280 >>> x = numpy.dtype('8f') 

6281 >>> x.base 

6282 dtype('float32') 

6283 

6284 >>> x = numpy.dtype('i2') 

6285 >>> x.base 

6286 dtype('int16') 

6287 

6288 """)) 

6289 

6290add_newdoc('numpy._core.multiarray', 'dtype', ('type', 

6291 """The type object used to instantiate a scalar of this data-type.""")) 

6292 

6293############################################################################## 

6294# 

6295# dtype methods 

6296# 

6297############################################################################## 

6298 

6299add_newdoc('numpy._core.multiarray', 'dtype', ('newbyteorder', 

6300 """ 

6301 newbyteorder(new_order='S', /) 

6302 

6303 Return a new dtype with a different byte order. 

6304 

6305 Changes are also made in all fields and sub-arrays of the data type. 

6306 

6307 Parameters 

6308 ---------- 

6309 new_order : string, optional 

6310 Byte order to force; a value from the byte order specifications 

6311 below. The default value ('S') results in swapping the current 

6312 byte order. `new_order` codes can be any of: 

6313 

6314 * 'S' - swap dtype from current to opposite endian 

6315 * {'<', 'little'} - little endian 

6316 * {'>', 'big'} - big endian 

6317 * {'=', 'native'} - native order 

6318 * {'|', 'I'} - ignore (no change to byte order) 

6319 

6320 Returns 

6321 ------- 

6322 new_dtype : dtype 

6323 New dtype object with the given change to the byte order. 

6324 

6325 Notes 

6326 ----- 

6327 Changes are also made in all fields and sub-arrays of the data type. 

6328 

6329 Examples 

6330 -------- 

6331 >>> import sys 

6332 >>> sys_is_le = sys.byteorder == 'little' 

6333 >>> native_code = '<' if sys_is_le else '>' 

6334 >>> swapped_code = '>' if sys_is_le else '<' 

6335 >>> import numpy as np 

6336 >>> native_dt = np.dtype(native_code+'i2') 

6337 >>> swapped_dt = np.dtype(swapped_code+'i2') 

6338 >>> native_dt.newbyteorder('S') == swapped_dt 

6339 True 

6340 >>> native_dt.newbyteorder() == swapped_dt 

6341 True 

6342 >>> native_dt == swapped_dt.newbyteorder('S') 

6343 True 

6344 >>> native_dt == swapped_dt.newbyteorder('=') 

6345 True 

6346 >>> native_dt == swapped_dt.newbyteorder('N') 

6347 True 

6348 >>> native_dt == native_dt.newbyteorder('|') 

6349 True 

6350 >>> np.dtype('<i2') == native_dt.newbyteorder('<') 

6351 True 

6352 >>> np.dtype('<i2') == native_dt.newbyteorder('L') 

6353 True 

6354 >>> np.dtype('>i2') == native_dt.newbyteorder('>') 

6355 True 

6356 >>> np.dtype('>i2') == native_dt.newbyteorder('B') 

6357 True 

6358 

6359 """)) 

6360 

6361add_newdoc('numpy._core.multiarray', 'dtype', ('__class_getitem__', 

6362 """ 

6363 __class_getitem__(item, /) 

6364 

6365 Return a parametrized wrapper around the `~numpy.dtype` type. 

6366 

6367 .. versionadded:: 1.22 

6368 

6369 Returns 

6370 ------- 

6371 alias : types.GenericAlias 

6372 A parametrized `~numpy.dtype` type. 

6373 

6374 Examples 

6375 -------- 

6376 >>> import numpy as np 

6377 

6378 >>> np.dtype[np.int64] 

6379 numpy.dtype[numpy.int64] 

6380 

6381 See Also 

6382 -------- 

6383 :pep:`585` : Type hinting generics in standard collections. 

6384 

6385 """)) 

6386 

6387add_newdoc('numpy._core.multiarray', 'dtype', ('__ge__', 

6388 """ 

6389 __ge__(value, /) 

6390 

6391 Return ``self >= value``. 

6392 

6393 Equivalent to ``np.can_cast(value, self, casting="safe")``. 

6394 

6395 See Also 

6396 -------- 

6397 can_cast : Returns True if cast between data types can occur according to 

6398 the casting rule. 

6399 

6400 """)) 

6401 

6402add_newdoc('numpy._core.multiarray', 'dtype', ('__le__', 

6403 """ 

6404 __le__(value, /) 

6405 

6406 Return ``self <= value``. 

6407 

6408 Equivalent to ``np.can_cast(self, value, casting="safe")``. 

6409 

6410 See Also 

6411 -------- 

6412 can_cast : Returns True if cast between data types can occur according to 

6413 the casting rule. 

6414 

6415 """)) 

6416 

6417add_newdoc('numpy._core.multiarray', 'dtype', ('__gt__', 

6418 """ 

6419 __ge__(value, /) 

6420 

6421 Return ``self > value``. 

6422 

6423 Equivalent to 

6424 ``self != value and np.can_cast(value, self, casting="safe")``. 

6425 

6426 See Also 

6427 -------- 

6428 can_cast : Returns True if cast between data types can occur according to 

6429 the casting rule. 

6430 

6431 """)) 

6432 

6433add_newdoc('numpy._core.multiarray', 'dtype', ('__lt__', 

6434 """ 

6435 __lt__(value, /) 

6436 

6437 Return ``self < value``. 

6438 

6439 Equivalent to 

6440 ``self != value and np.can_cast(self, value, casting="safe")``. 

6441 

6442 See Also 

6443 -------- 

6444 can_cast : Returns True if cast between data types can occur according to 

6445 the casting rule. 

6446 

6447 """)) 

6448 

6449############################################################################## 

6450# 

6451# Datetime-related Methods 

6452# 

6453############################################################################## 

6454 

6455add_newdoc('numpy._core.multiarray', 'busdaycalendar', 

6456 """ 

6457 busdaycalendar(weekmask='1111100', holidays=None) 

6458 

6459 A business day calendar object that efficiently stores information 

6460 defining valid days for the busday family of functions. 

6461 

6462 The default valid days are Monday through Friday ("business days"). 

6463 A busdaycalendar object can be specified with any set of weekly 

6464 valid days, plus an optional "holiday" dates that always will be invalid. 

6465 

6466 Once a busdaycalendar object is created, the weekmask and holidays 

6467 cannot be modified. 

6468 

6469 Parameters 

6470 ---------- 

6471 weekmask : str or array_like of bool, optional 

6472 A seven-element array indicating which of Monday through Sunday are 

6473 valid days. May be specified as a length-seven list or array, like 

6474 [1,1,1,1,1,0,0]; a length-seven string, like '1111100'; or a string 

6475 like "Mon Tue Wed Thu Fri", made up of 3-character abbreviations for 

6476 weekdays, optionally separated by white space. Valid abbreviations 

6477 are: Mon Tue Wed Thu Fri Sat Sun 

6478 holidays : array_like of datetime64[D], optional 

6479 An array of dates to consider as invalid dates, no matter which 

6480 weekday they fall upon. Holiday dates may be specified in any 

6481 order, and NaT (not-a-time) dates are ignored. This list is 

6482 saved in a normalized form that is suited for fast calculations 

6483 of valid days. 

6484 

6485 Returns 

6486 ------- 

6487 out : busdaycalendar 

6488 A business day calendar object containing the specified 

6489 weekmask and holidays values. 

6490 

6491 See Also 

6492 -------- 

6493 is_busday : Returns a boolean array indicating valid days. 

6494 busday_offset : Applies an offset counted in valid days. 

6495 busday_count : Counts how many valid days are in a half-open date range. 

6496 

6497 Attributes 

6498 ---------- 

6499 weekmask : (copy) seven-element array of bool 

6500 holidays : (copy) sorted array of datetime64[D] 

6501 

6502 Notes 

6503 ----- 

6504 Once a busdaycalendar object is created, you cannot modify the 

6505 weekmask or holidays. The attributes return copies of internal data. 

6506 

6507 Examples 

6508 -------- 

6509 >>> import numpy as np 

6510 >>> # Some important days in July 

6511 ... bdd = np.busdaycalendar( 

6512 ... holidays=['2011-07-01', '2011-07-04', '2011-07-17']) 

6513 >>> # Default is Monday to Friday weekdays 

6514 ... bdd.weekmask 

6515 array([ True, True, True, True, True, False, False]) 

6516 >>> # Any holidays already on the weekend are removed 

6517 ... bdd.holidays 

6518 array(['2011-07-01', '2011-07-04'], dtype='datetime64[D]') 

6519 """) 

6520 

6521add_newdoc('numpy._core.multiarray', 'busdaycalendar', ('weekmask', 

6522 """A copy of the seven-element boolean mask indicating valid days.""")) 

6523 

6524add_newdoc('numpy._core.multiarray', 'busdaycalendar', ('holidays', 

6525 """A copy of the holiday array indicating additional invalid days.""")) 

6526 

6527add_newdoc('numpy._core.multiarray', 'normalize_axis_index', 

6528 """ 

6529 normalize_axis_index(axis, ndim, msg_prefix=None) 

6530 

6531 Normalizes an axis index, `axis`, such that is a valid positive index into 

6532 the shape of array with `ndim` dimensions. Raises an AxisError with an 

6533 appropriate message if this is not possible. 

6534 

6535 Used internally by all axis-checking logic. 

6536 

6537 Parameters 

6538 ---------- 

6539 axis : int 

6540 The un-normalized index of the axis. Can be negative 

6541 ndim : int 

6542 The number of dimensions of the array that `axis` should be normalized 

6543 against 

6544 msg_prefix : str 

6545 A prefix to put before the message, typically the name of the argument 

6546 

6547 Returns 

6548 ------- 

6549 normalized_axis : int 

6550 The normalized axis index, such that `0 <= normalized_axis < ndim` 

6551 

6552 Raises 

6553 ------ 

6554 AxisError 

6555 If the axis index is invalid, when `-ndim <= axis < ndim` is false. 

6556 

6557 Examples 

6558 -------- 

6559 >>> import numpy as np 

6560 >>> from numpy.lib.array_utils import normalize_axis_index 

6561 >>> normalize_axis_index(0, ndim=3) 

6562 0 

6563 >>> normalize_axis_index(1, ndim=3) 

6564 1 

6565 >>> normalize_axis_index(-1, ndim=3) 

6566 2 

6567 

6568 >>> normalize_axis_index(3, ndim=3) 

6569 Traceback (most recent call last): 

6570 ... 

6571 numpy.exceptions.AxisError: axis 3 is out of bounds for array ... 

6572 >>> normalize_axis_index(-4, ndim=3, msg_prefix='axes_arg') 

6573 Traceback (most recent call last): 

6574 ... 

6575 numpy.exceptions.AxisError: axes_arg: axis -4 is out of bounds ... 

6576 """) 

6577 

6578add_newdoc('numpy._core.multiarray', 'datetime_data', 

6579 """ 

6580 datetime_data(dtype, /) 

6581 

6582 Get information about the step size of a date or time type. 

6583 

6584 The returned tuple can be passed as the second argument of `numpy.datetime64` and 

6585 `numpy.timedelta64`. 

6586 

6587 Parameters 

6588 ---------- 

6589 dtype : dtype 

6590 The dtype object, which must be a `datetime64` or `timedelta64` type. 

6591 

6592 Returns 

6593 ------- 

6594 unit : str 

6595 The :ref:`datetime unit <arrays.dtypes.dateunits>` on which this dtype 

6596 is based. 

6597 count : int 

6598 The number of base units in a step. 

6599 

6600 Examples 

6601 -------- 

6602 >>> import numpy as np 

6603 >>> dt_25s = np.dtype('timedelta64[25s]') 

6604 >>> np.datetime_data(dt_25s) 

6605 ('s', 25) 

6606 >>> np.array(10, dt_25s).astype('timedelta64[s]') 

6607 array(250, dtype='timedelta64[s]') 

6608 

6609 The result can be used to construct a datetime that uses the same units 

6610 as a timedelta 

6611 

6612 >>> np.datetime64('2010', np.datetime_data(dt_25s)) 

6613 np.datetime64('2010-01-01T00:00:00','25s') 

6614 """) 

6615 

6616 

6617############################################################################## 

6618# 

6619# Documentation for `generic` attributes and methods 

6620# 

6621############################################################################## 

6622 

6623add_newdoc('numpy._core.numerictypes', 'generic', 

6624 """ 

6625 Base class for numpy scalar types. 

6626 

6627 Class from which most (all?) numpy scalar types are derived. For 

6628 consistency, exposes the same API as `ndarray`, despite many 

6629 consequent attributes being either "get-only," or completely irrelevant. 

6630 This is the class from which it is strongly suggested users should derive 

6631 custom scalar types. 

6632 

6633 """) 

6634 

6635# Attributes 

6636 

6637def refer_to_array_attribute(attr, method=True): 

6638 docstring = """ 

6639 Scalar {} identical to the corresponding array attribute. 

6640 

6641 Please see `ndarray.{}`. 

6642 """ 

6643 

6644 return attr, docstring.format("method" if method else "attribute", attr) 

6645 

6646 

6647add_newdoc('numpy._core.numerictypes', 'generic', 

6648 refer_to_array_attribute('T', method=False)) 

6649 

6650add_newdoc('numpy._core.numerictypes', 'generic', 

6651 refer_to_array_attribute('base', method=False)) 

6652 

6653add_newdoc('numpy._core.numerictypes', 'generic', ('data', 

6654 """Pointer to start of data.""")) 

6655 

6656add_newdoc('numpy._core.numerictypes', 'generic', ('dtype', 

6657 """Get array data-descriptor.""")) 

6658 

6659add_newdoc('numpy._core.numerictypes', 'generic', ('flags', 

6660 """The integer value of flags.""")) 

6661 

6662add_newdoc('numpy._core.numerictypes', 'generic', ('flat', 

6663 """A 1-D view of the scalar.""")) 

6664 

6665add_newdoc('numpy._core.numerictypes', 'generic', ('imag', 

6666 """The imaginary part of the scalar.""")) 

6667 

6668add_newdoc('numpy._core.numerictypes', 'generic', ('itemsize', 

6669 """The length of one element in bytes.""")) 

6670 

6671add_newdoc('numpy._core.numerictypes', 'generic', ('ndim', 

6672 """The number of array dimensions.""")) 

6673 

6674add_newdoc('numpy._core.numerictypes', 'generic', ('real', 

6675 """The real part of the scalar.""")) 

6676 

6677add_newdoc('numpy._core.numerictypes', 'generic', ('shape', 

6678 """Tuple of array dimensions.""")) 

6679 

6680add_newdoc('numpy._core.numerictypes', 'generic', ('size', 

6681 """The number of elements in the gentype.""")) 

6682 

6683add_newdoc('numpy._core.numerictypes', 'generic', ('strides', 

6684 """Tuple of bytes steps in each dimension.""")) 

6685 

6686# Methods 

6687 

6688add_newdoc('numpy._core.numerictypes', 'generic', 

6689 refer_to_array_attribute('all')) 

6690 

6691add_newdoc('numpy._core.numerictypes', 'generic', 

6692 refer_to_array_attribute('any')) 

6693 

6694add_newdoc('numpy._core.numerictypes', 'generic', 

6695 refer_to_array_attribute('argmax')) 

6696 

6697add_newdoc('numpy._core.numerictypes', 'generic', 

6698 refer_to_array_attribute('argmin')) 

6699 

6700add_newdoc('numpy._core.numerictypes', 'generic', 

6701 refer_to_array_attribute('argsort')) 

6702 

6703add_newdoc('numpy._core.numerictypes', 'generic', 

6704 refer_to_array_attribute('astype')) 

6705 

6706add_newdoc('numpy._core.numerictypes', 'generic', 

6707 refer_to_array_attribute('byteswap')) 

6708 

6709add_newdoc('numpy._core.numerictypes', 'generic', 

6710 refer_to_array_attribute('choose')) 

6711 

6712add_newdoc('numpy._core.numerictypes', 'generic', 

6713 refer_to_array_attribute('clip')) 

6714 

6715add_newdoc('numpy._core.numerictypes', 'generic', 

6716 refer_to_array_attribute('compress')) 

6717 

6718add_newdoc('numpy._core.numerictypes', 'generic', 

6719 refer_to_array_attribute('conjugate')) 

6720 

6721add_newdoc('numpy._core.numerictypes', 'generic', 

6722 refer_to_array_attribute('copy')) 

6723 

6724add_newdoc('numpy._core.numerictypes', 'generic', 

6725 refer_to_array_attribute('cumprod')) 

6726 

6727add_newdoc('numpy._core.numerictypes', 'generic', 

6728 refer_to_array_attribute('cumsum')) 

6729 

6730add_newdoc('numpy._core.numerictypes', 'generic', 

6731 refer_to_array_attribute('diagonal')) 

6732 

6733add_newdoc('numpy._core.numerictypes', 'generic', 

6734 refer_to_array_attribute('dump')) 

6735 

6736add_newdoc('numpy._core.numerictypes', 'generic', 

6737 refer_to_array_attribute('dumps')) 

6738 

6739add_newdoc('numpy._core.numerictypes', 'generic', 

6740 refer_to_array_attribute('fill')) 

6741 

6742add_newdoc('numpy._core.numerictypes', 'generic', 

6743 refer_to_array_attribute('flatten')) 

6744 

6745add_newdoc('numpy._core.numerictypes', 'generic', 

6746 refer_to_array_attribute('getfield')) 

6747 

6748add_newdoc('numpy._core.numerictypes', 'generic', 

6749 refer_to_array_attribute('item')) 

6750 

6751add_newdoc('numpy._core.numerictypes', 'generic', 

6752 refer_to_array_attribute('max')) 

6753 

6754add_newdoc('numpy._core.numerictypes', 'generic', 

6755 refer_to_array_attribute('mean')) 

6756 

6757add_newdoc('numpy._core.numerictypes', 'generic', 

6758 refer_to_array_attribute('min')) 

6759 

6760add_newdoc('numpy._core.numerictypes', 'generic', 

6761 refer_to_array_attribute('nonzero')) 

6762 

6763add_newdoc('numpy._core.numerictypes', 'generic', 

6764 refer_to_array_attribute('prod')) 

6765 

6766add_newdoc('numpy._core.numerictypes', 'generic', 

6767 refer_to_array_attribute('put')) 

6768 

6769add_newdoc('numpy._core.numerictypes', 'generic', 

6770 refer_to_array_attribute('ravel')) 

6771 

6772add_newdoc('numpy._core.numerictypes', 'generic', 

6773 refer_to_array_attribute('repeat')) 

6774 

6775add_newdoc('numpy._core.numerictypes', 'generic', 

6776 refer_to_array_attribute('reshape')) 

6777 

6778add_newdoc('numpy._core.numerictypes', 'generic', 

6779 refer_to_array_attribute('resize')) 

6780 

6781add_newdoc('numpy._core.numerictypes', 'generic', 

6782 refer_to_array_attribute('round')) 

6783 

6784add_newdoc('numpy._core.numerictypes', 'generic', 

6785 refer_to_array_attribute('searchsorted')) 

6786 

6787add_newdoc('numpy._core.numerictypes', 'generic', 

6788 refer_to_array_attribute('setfield')) 

6789 

6790add_newdoc('numpy._core.numerictypes', 'generic', 

6791 refer_to_array_attribute('setflags')) 

6792 

6793add_newdoc('numpy._core.numerictypes', 'generic', 

6794 refer_to_array_attribute('sort')) 

6795 

6796add_newdoc('numpy._core.numerictypes', 'generic', 

6797 refer_to_array_attribute('squeeze')) 

6798 

6799add_newdoc('numpy._core.numerictypes', 'generic', 

6800 refer_to_array_attribute('std')) 

6801 

6802add_newdoc('numpy._core.numerictypes', 'generic', 

6803 refer_to_array_attribute('sum')) 

6804 

6805add_newdoc('numpy._core.numerictypes', 'generic', 

6806 refer_to_array_attribute('swapaxes')) 

6807 

6808add_newdoc('numpy._core.numerictypes', 'generic', 

6809 refer_to_array_attribute('take')) 

6810 

6811add_newdoc('numpy._core.numerictypes', 'generic', 

6812 refer_to_array_attribute('tofile')) 

6813 

6814add_newdoc('numpy._core.numerictypes', 'generic', 

6815 refer_to_array_attribute('tolist')) 

6816 

6817add_newdoc('numpy._core.numerictypes', 'generic', 

6818 refer_to_array_attribute('tostring')) 

6819 

6820add_newdoc('numpy._core.numerictypes', 'generic', 

6821 refer_to_array_attribute('trace')) 

6822 

6823add_newdoc('numpy._core.numerictypes', 'generic', 

6824 refer_to_array_attribute('transpose')) 

6825 

6826add_newdoc('numpy._core.numerictypes', 'generic', 

6827 refer_to_array_attribute('var')) 

6828 

6829add_newdoc('numpy._core.numerictypes', 'generic', 

6830 refer_to_array_attribute('view')) 

6831 

6832add_newdoc('numpy._core.numerictypes', 'number', ('__class_getitem__', 

6833 """ 

6834 __class_getitem__(item, /) 

6835 

6836 Return a parametrized wrapper around the `~numpy.number` type. 

6837 

6838 .. versionadded:: 1.22 

6839 

6840 Returns 

6841 ------- 

6842 alias : types.GenericAlias 

6843 A parametrized `~numpy.number` type. 

6844 

6845 Examples 

6846 -------- 

6847 >>> from typing import Any 

6848 >>> import numpy as np 

6849 

6850 >>> np.signedinteger[Any] 

6851 numpy.signedinteger[typing.Any] 

6852 

6853 See Also 

6854 -------- 

6855 :pep:`585` : Type hinting generics in standard collections. 

6856 

6857 """)) 

6858 

6859############################################################################## 

6860# 

6861# Documentation for scalar type abstract base classes in type hierarchy 

6862# 

6863############################################################################## 

6864 

6865 

6866add_newdoc('numpy._core.numerictypes', 'number', 

6867 """ 

6868 Abstract base class of all numeric scalar types. 

6869 

6870 """) 

6871 

6872add_newdoc('numpy._core.numerictypes', 'integer', 

6873 """ 

6874 Abstract base class of all integer scalar types. 

6875 

6876 """) 

6877 

6878add_newdoc('numpy._core.numerictypes', 'signedinteger', 

6879 """ 

6880 Abstract base class of all signed integer scalar types. 

6881 

6882 """) 

6883 

6884add_newdoc('numpy._core.numerictypes', 'unsignedinteger', 

6885 """ 

6886 Abstract base class of all unsigned integer scalar types. 

6887 

6888 """) 

6889 

6890add_newdoc('numpy._core.numerictypes', 'inexact', 

6891 """ 

6892 Abstract base class of all numeric scalar types with a (potentially) 

6893 inexact representation of the values in its range, such as 

6894 floating-point numbers. 

6895 

6896 """) 

6897 

6898add_newdoc('numpy._core.numerictypes', 'floating', 

6899 """ 

6900 Abstract base class of all floating-point scalar types. 

6901 

6902 """) 

6903 

6904add_newdoc('numpy._core.numerictypes', 'complexfloating', 

6905 """ 

6906 Abstract base class of all complex number scalar types that are made up of 

6907 floating-point numbers. 

6908 

6909 """) 

6910 

6911add_newdoc('numpy._core.numerictypes', 'flexible', 

6912 """ 

6913 Abstract base class of all scalar types without predefined length. 

6914 The actual size of these types depends on the specific `numpy.dtype` 

6915 instantiation. 

6916 

6917 """) 

6918 

6919add_newdoc('numpy._core.numerictypes', 'character', 

6920 """ 

6921 Abstract base class of all character string scalar types. 

6922 

6923 """) 

6924 

6925add_newdoc('numpy._core.multiarray', 'StringDType', 

6926 """ 

6927 StringDType(*, na_object=np._NoValue, coerce=True) 

6928 

6929 Create a StringDType instance. 

6930 

6931 StringDType can be used to store UTF-8 encoded variable-width strings in 

6932 a NumPy array. 

6933 

6934 Parameters 

6935 ---------- 

6936 na_object : object, optional 

6937 Object used to represent missing data. If unset, the array will not 

6938 use a missing data sentinel. 

6939 coerce : bool, optional 

6940 Whether or not items in an array-like passed to an array creation 

6941 function that are neither a str or str subtype should be coerced to 

6942 str. Defaults to True. If set to False, creating a StringDType 

6943 array from an array-like containing entries that are not already 

6944 strings will raise an error. 

6945 

6946 Examples 

6947 -------- 

6948 

6949 >>> import numpy as np 

6950 

6951 >>> from numpy.dtypes import StringDType 

6952 >>> np.array(["hello", "world"], dtype=StringDType()) 

6953 array(["hello", "world"], dtype=StringDType()) 

6954 

6955 >>> arr = np.array(["hello", None, "world"], 

6956 ... dtype=StringDType(na_object=None)) 

6957 >>> arr 

6958 array(["hello", None, "world"], dtype=StringDType(na_object=None)) 

6959 >>> arr[1] is None 

6960 True 

6961 

6962 >>> arr = np.array(["hello", np.nan, "world"], 

6963 ... dtype=StringDType(na_object=np.nan)) 

6964 >>> np.isnan(arr) 

6965 array([False, True, False]) 

6966 

6967 >>> np.array([1.2, object(), "hello world"], 

6968 ... dtype=StringDType(coerce=True)) 

6969 ValueError: StringDType only allows string data when string coercion 

6970 is disabled. 

6971 

6972 >>> np.array(["hello", "world"], dtype=StringDType(coerce=True)) 

6973 array(["hello", "world"], dtype=StringDType(coerce=True)) 

6974 """)