Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/tensorflow/python/util/nest.py: 51%
122 statements
« prev ^ index » next coverage.py v7.4.0, created at 2024-01-03 07:57 +0000
« prev ^ index » next coverage.py v7.4.0, created at 2024-01-03 07:57 +0000
1# Copyright 2021 The TensorFlow Authors. All Rights Reserved.
2#
3# Licensed under the Apache License, Version 2.0 (the "License");
4# you may not use this file except in compliance with the License.
5# You may obtain a copy of the License at
6#
7# http://www.apache.org/licenses/LICENSE-2.0
8#
9# Unless required by applicable law or agreed to in writing, software
10# distributed under the License is distributed on an "AS IS" BASIS,
11# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12# See the License for the specific language governing permissions and
13# limitations under the License.
14# ==============================================================================
16"""Functions that work with structures.
18A structure is either:
20* one of the recognized Python collections, holding _nested structures_;
21* a value of any other type, typically a TensorFlow data type like Tensor,
22 Variable, or of compatible types such as int, float, ndarray, etc. these are
23 commonly referred to as _atoms_ of the structure.
25A structure of type `T` is a structure whose atomic items are of type `T`.
26For example, a structure of `tf.Tensor` only contains `tf.Tensor` as its atoms.
28Historically a _nested structure_ was called a _nested sequence_ in TensorFlow.
29A nested structure is sometimes called a _nest_ or a _tree_, but the formal
30name _nested structure_ is preferred.
32Refer to [Nesting Data Structures]
33(https://en.wikipedia.org/wiki/Nesting_(computing)#Data_structures).
35The following collection types are recognized by `tf.nest` as nested
36structures:
38* `collections.abc.Sequence` (except `string` and `bytes`).
39 This includes `list`, `tuple`, and `namedtuple`.
40* `collections.abc.Mapping` (with sortable keys).
41 This includes `dict` and `collections.OrderedDict`.
42* `collections.abc.MappingView` (with sortable keys).
43* [`attr.s` classes](https://www.attrs.org/).
45Any other values are considered **atoms**. Not all collection types are
46considered nested structures. For example, the following types are
47considered atoms:
49* `set`; `{"a", "b"}` is an atom, while `["a", "b"]` is a nested structure.
50* [`dataclass` classes](https://docs.python.org/library/dataclasses.html)
51* `tf.Tensor`
52* `numpy.array`
54`tf.nest.is_nested` checks whether an object is a nested structure or an atom.
55For example:
57 >>> tf.nest.is_nested("1234")
58 False
59 >>> tf.nest.is_nested([1, 3, [4, 5]])
60 True
61 >>> tf.nest.is_nested(((7, 8), (5, 6)))
62 True
63 >>> tf.nest.is_nested([])
64 True
65 >>> tf.nest.is_nested({"a": 1, "b": 2})
66 True
67 >>> tf.nest.is_nested({"a": 1, "b": 2}.keys())
68 True
69 >>> tf.nest.is_nested({"a": 1, "b": 2}.values())
70 True
71 >>> tf.nest.is_nested({"a": 1, "b": 2}.items())
72 True
73 >>> tf.nest.is_nested(set([1, 2]))
74 False
75 >>> ones = tf.ones([2, 3])
76 >>> tf.nest.is_nested(ones)
77 False
79Note: A proper structure shall form a tree. The user shall ensure there is no
80cyclic references within the items in the structure,
81i.e., no references in the structure of the input of these functions
82should be recursive. The behavior is undefined if there is a cycle.
84"""
86import wrapt as _wrapt
88from tensorflow.python.util import _pywrap_nest
89from tensorflow.python.util import _pywrap_utils
90from tensorflow.python.util import nest_util
91from tensorflow.python.util.compat import collections_abc as _collections_abc
92from tensorflow.python.util.tf_export import tf_export
95STRUCTURES_HAVE_MISMATCHING_LENGTHS = (
96 nest_util.STRUCTURES_HAVE_MISMATCHING_LENGTHS
97)
99STRUCTURES_HAVE_MISMATCHING_TYPES = nest_util.STRUCTURES_HAVE_MISMATCHING_TYPES
101SHALLOW_TREE_HAS_INVALID_KEYS = nest_util.SHALLOW_TREE_HAS_INVALID_KEYS
103INPUT_TREE_SMALLER_THAN_SHALLOW_TREE = (
104 nest_util.INPUT_TREE_SMALLER_THAN_SHALLOW_TREE
105)
107IF_SHALLOW_IS_SEQ_INPUT_MUST_BE_SEQ = (
108 "If shallow structure is a sequence, input must also be a sequence. "
109 "Input has type: {}."
110)
112is_namedtuple = nest_util.is_namedtuple
113_is_namedtuple = nest_util.is_namedtuple
114_is_attrs = _pywrap_utils.IsAttrs
115_is_mapping = _pywrap_utils.IsMapping
116same_namedtuples = nest_util.same_namedtuples
119def _yield_value(iterable):
120 return nest_util.yield_value(nest_util.Modality.CORE, iterable)
123def _yield_sorted_items(iterable):
124 return nest_util.yield_sorted_items(nest_util.Modality.CORE, iterable)
127@tf_export("__internal__.nest.is_mapping", v1=[])
128def is_mapping(obj):
129 """Returns a true if its input is a collections.Mapping."""
130 return _is_mapping(obj)
133# TODO(b/225045380): Move to a "leaf" library to use in trace_type.
134@tf_export("__internal__.nest.is_attrs", v1=[])
135def is_attrs(obj):
136 """Returns a true if its input is an instance of an attr.s decorated class."""
137 return _is_attrs(obj)
140@tf_export("__internal__.nest.sequence_like", v1=[])
141def _sequence_like(instance, args):
142 """Converts the sequence `args` to the same type as `instance`.
144 Args:
145 instance: an instance of `tuple`, `list`, `namedtuple`, `dict`,
146 `collections.OrderedDict`, or `composite_tensor.Composite_Tensor`
147 or `type_spec.TypeSpec`.
148 args: items to be converted to the `instance` type.
150 Returns:
151 `args` with the type of `instance`.
152 """
153 return nest_util.sequence_like(instance, args)
156_is_nested_or_composite = _pywrap_utils.IsNestedOrComposite
159@tf_export("nest.is_nested")
160def is_nested(seq):
161 """Returns true if its input is a nested structure.
163 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
164 for the definition of a nested structure.
166 Args:
167 seq: the value to test.
169 Returns:
170 True if the input is a nested structure.
171 """
172 return nest_util.is_nested(nest_util.Modality.CORE, seq)
175def is_nested_or_composite(seq):
176 """Returns true if its input is a nested structure or a composite.
178 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
179 for the definition of a nested structure.
181 Args:
182 seq: the value to test.
184 Returns:
185 True if the input is a nested structure or a composite.
186 """
187 return _is_nested_or_composite(seq)
190def is_sequence_or_composite(seq):
191 return _is_nested_or_composite(seq)
194@tf_export("nest.flatten")
195def flatten(structure, expand_composites=False):
196 """Returns a flat list from a given structure.
198 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
199 for the definition of a structure.
201 If the structure is an atom, then returns a single-item list: [structure].
203 This is the inverse of the `nest.pack_sequence_as` method that takes in a
204 flattened list and re-packs it into the nested structure.
206 In the case of dict instances, the sequence consists of the values, sorted by
207 key to ensure deterministic behavior. This is true also for OrderedDict
208 instances: their sequence order is ignored, the sorting order of keys is used
209 instead. The same convention is followed in `nest.pack_sequence_as`. This
210 correctly repacks dicts and OrderedDicts after they have been flattened, and
211 also allows flattening an OrderedDict and then repacking it back using a
212 corresponding plain dict, or vice-versa. Dictionaries with non-sortable keys
213 cannot be flattened.
215 Users must not modify any collections used in nest while this function is
216 running.
218 Examples:
220 1. Python dict (ordered by key):
222 >>> dict = { "key3": "value3", "key1": "value1", "key2": "value2" }
223 >>> tf.nest.flatten(dict)
224 ['value1', 'value2', 'value3']
226 2. For a nested python tuple:
228 >>> tuple = ((1.0, 2.0), (3.0, 4.0, 5.0), 6.0)
229 >>> tf.nest.flatten(tuple)
230 [1.0, 2.0, 3.0, 4.0, 5.0, 6.0]
232 3. For a nested dictionary of dictionaries:
234 >>> dict = { "key3": {"c": (1.0, 2.0), "a": (3.0)},
235 ... "key1": {"m": "val1", "g": "val2"} }
236 >>> tf.nest.flatten(dict)
237 ['val2', 'val1', 3.0, 1.0, 2.0]
239 4. Numpy array (will not flatten):
241 >>> array = np.array([[1, 2], [3, 4]])
242 >>> tf.nest.flatten(array)
243 [array([[1, 2],
244 [3, 4]])]
246 5. `tf.Tensor` (will not flatten):
248 >>> tensor = tf.constant([[1., 2., 3.], [4., 5., 6.], [7., 8., 9.]])
249 >>> tf.nest.flatten(tensor)
250 [<tf.Tensor: shape=(3, 3), dtype=float32, numpy=
251 array([[1., 2., 3.],
252 [4., 5., 6.],
253 [7., 8., 9.]], dtype=float32)>]
255 6. `tf.RaggedTensor`: This is a composite tensor thats representation consists
256 of a flattened list of 'values' and a list of 'row_splits' which indicate how
257 to chop up the flattened list into different rows. For more details on
258 `tf.RaggedTensor`, please visit
259 https://www.tensorflow.org/api_docs/python/tf/RaggedTensor.
261 with `expand_composites=False`, we just return the RaggedTensor as is.
263 >>> tensor = tf.ragged.constant([[3, 1, 4, 1], [], [5, 9, 2]])
264 >>> tf.nest.flatten(tensor, expand_composites=False)
265 [<tf.RaggedTensor [[3, 1, 4, 1], [], [5, 9, 2]]>]
267 with `expand_composites=True`, we return the component Tensors that make up
268 the RaggedTensor representation (the values and row_splits tensors)
270 >>> tensor = tf.ragged.constant([[3, 1, 4, 1], [], [5, 9, 2]])
271 >>> tf.nest.flatten(tensor, expand_composites=True)
272 [<tf.Tensor: shape=(7,), dtype=int32, numpy=array([3, 1, 4, 1, 5, 9, 2],
273 dtype=int32)>,
274 <tf.Tensor: shape=(4,), dtype=int64, numpy=array([0, 4, 4, 7])>]
276 Args:
277 structure: an atom or a nested structure. Note, numpy arrays are considered
278 atoms and are not flattened.
279 expand_composites: If true, then composite tensors such as
280 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
281 component tensors.
283 Returns:
284 A Python list, the flattened version of the input.
286 Raises:
287 TypeError: The nest is or contains a dict with non-sortable keys.
288 """
289 return nest_util.flatten(
290 nest_util.Modality.CORE, structure, expand_composites
291 )
294@tf_export("nest.assert_same_structure")
295def assert_same_structure(nest1, nest2, check_types=True,
296 expand_composites=False):
297 """Asserts that two structures are nested in the same way.
299 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
300 for the definition of a structure.
302 Note the method does not check the types of atoms inside the structures.
304 Examples:
306 * These atom vs. atom comparisons will pass:
308 >>> tf.nest.assert_same_structure(1.5, tf.Variable(1, tf.uint32))
309 >>> tf.nest.assert_same_structure("abc", np.array([1, 2]))
311 * These nested structure vs. nested structure comparisons will pass:
313 >>> structure1 = (((1, 2), 3), 4, (5, 6))
314 >>> structure2 = ((("foo1", "foo2"), "foo3"), "foo4", ("foo5", "foo6"))
315 >>> structure3 = [(("a", "b"), "c"), "d", ["e", "f"]]
316 >>> tf.nest.assert_same_structure(structure1, structure2)
317 >>> tf.nest.assert_same_structure(structure1, structure3, check_types=False)
319 >>> import collections
320 >>> tf.nest.assert_same_structure(
321 ... collections.namedtuple("bar", "a b")(1, 2),
322 ... collections.namedtuple("foo", "a b")(2, 3),
323 ... check_types=False)
325 >>> tf.nest.assert_same_structure(
326 ... collections.namedtuple("bar", "a b")(1, 2),
327 ... { "a": 1, "b": 2 },
328 ... check_types=False)
330 >>> tf.nest.assert_same_structure(
331 ... { "a": 1, "b": 2, "c": 3 },
332 ... { "c": 6, "b": 5, "a": 4 })
334 >>> ragged_tensor1 = tf.RaggedTensor.from_row_splits(
335 ... values=[3, 1, 4, 1, 5, 9, 2, 6],
336 ... row_splits=[0, 4, 4, 7, 8, 8])
337 >>> ragged_tensor2 = tf.RaggedTensor.from_row_splits(
338 ... values=[3, 1, 4],
339 ... row_splits=[0, 3])
340 >>> tf.nest.assert_same_structure(
341 ... ragged_tensor1,
342 ... ragged_tensor2,
343 ... expand_composites=True)
345 * These examples will raise exceptions:
347 >>> tf.nest.assert_same_structure([0, 1], np.array([0, 1]))
348 Traceback (most recent call last):
349 ...
350 ValueError: The two structures don't have the same nested structure
352 >>> tf.nest.assert_same_structure(
353 ... collections.namedtuple('bar', 'a b')(1, 2),
354 ... collections.namedtuple('foo', 'a b')(2, 3))
355 Traceback (most recent call last):
356 ...
357 TypeError: The two structures don't have the same nested structure
359 Args:
360 nest1: an atom or a nested structure.
361 nest2: an atom or a nested structure.
362 check_types: if `True` (default) types of structures are checked as well,
363 including the keys of dictionaries. If set to `False`, for example a list
364 and a tuple of objects will look the same if they have the same size. Note
365 that namedtuples with identical name and fields are always considered to
366 have the same shallow structure. Two types will also be considered the
367 same if they are both list subtypes (which allows "list" and
368 "_ListWrapper" from trackable dependency tracking to compare equal).
369 `check_types=True` only checks type of sub-structures. The types of atoms
370 are not checked.
371 expand_composites: If true, then composite tensors such as
372 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
373 component tensors.
375 Raises:
376 ValueError: If the two structures do not have the same number of atoms or
377 if the two structures are not nested in the same way.
378 TypeError: If the two structures differ in the type of sequence in any of
379 their substructures. Only possible if `check_types` is `True`.
380 """
381 nest_util.assert_same_structure(
382 nest_util.Modality.CORE, nest1, nest2, check_types, expand_composites
383 )
386def flatten_dict_items(dictionary):
387 """Returns a dictionary with flattened keys and values.
389 This function flattens the keys and values of a dictionary, which can be
390 arbitrarily nested structures, and returns the flattened version of such
391 structures:
393 ```python
394 example_dictionary = {(4, 5, (6, 8)): ("a", "b", ("c", "d"))}
395 result = {4: "a", 5: "b", 6: "c", 8: "d"}
396 flatten_dict_items(example_dictionary) == result
397 ```
399 The input dictionary must satisfy two properties:
401 1. Its keys and values should have the same exact nested structure.
402 2. The set of all flattened keys of the dictionary must not contain repeated
403 keys.
405 Args:
406 dictionary: the dictionary to zip
408 Returns:
409 The zipped dictionary.
411 Raises:
412 TypeError: If the input is not a dictionary.
413 ValueError: If any key and value do not have the same structure layout, or
414 if keys are not unique.
415 """
416 return _pywrap_nest.FlattenDictItems(dictionary)
419@tf_export("nest.pack_sequence_as")
420def pack_sequence_as(structure, flat_sequence, expand_composites=False):
421 """Returns a given flattened sequence packed into a given structure.
423 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
424 for the definition of a structure.
426 If `structure` is an atom, `flat_sequence` must be a single-item list;
427 in this case the return value is `flat_sequence[0]`.
429 If `structure` is or contains a dict instance, the keys will be sorted to
430 pack the flat sequence in deterministic order. This is true also for
431 `OrderedDict` instances: their sequence order is ignored, the sorting order of
432 keys is used instead. The same convention is followed in `flatten`.
433 This correctly repacks dicts and `OrderedDict`s after they have been
434 flattened, and also allows flattening an `OrderedDict` and then repacking it
435 back using a corresponding plain dict, or vice-versa.
436 Dictionaries with non-sortable keys cannot be flattened.
438 Examples:
440 1. Python dict:
442 >>> structure = { "key3": "", "key1": "", "key2": "" }
443 >>> flat_sequence = ["value1", "value2", "value3"]
444 >>> tf.nest.pack_sequence_as(structure, flat_sequence)
445 {'key3': 'value3', 'key1': 'value1', 'key2': 'value2'}
447 2. For a nested python tuple:
449 >>> structure = (('a','b'), ('c','d','e'), 'f')
450 >>> flat_sequence = [1.0, 2.0, 3.0, 4.0, 5.0, 6.0]
451 >>> tf.nest.pack_sequence_as(structure, flat_sequence)
452 ((1.0, 2.0), (3.0, 4.0, 5.0), 6.0)
454 3. For a nested dictionary of dictionaries:
456 >>> structure = { "key3": {"c": ('alpha', 'beta'), "a": ('gamma')},
457 ... "key1": {"e": "val1", "d": "val2"} }
458 >>> flat_sequence = ['val2', 'val1', 3.0, 1.0, 2.0]
459 >>> tf.nest.pack_sequence_as(structure, flat_sequence)
460 {'key3': {'c': (1.0, 2.0), 'a': 3.0}, 'key1': {'e': 'val1', 'd': 'val2'}}
462 4. Numpy array (considered a scalar):
464 >>> structure = ['a']
465 >>> flat_sequence = [np.array([[1, 2], [3, 4]])]
466 >>> tf.nest.pack_sequence_as(structure, flat_sequence)
467 [array([[1, 2],
468 [3, 4]])]
470 5. tf.Tensor (considered a scalar):
472 >>> structure = ['a']
473 >>> flat_sequence = [tf.constant([[1., 2., 3.], [4., 5., 6.]])]
474 >>> tf.nest.pack_sequence_as(structure, flat_sequence)
475 [<tf.Tensor: shape=(2, 3), dtype=float32,
476 numpy= array([[1., 2., 3.], [4., 5., 6.]], dtype=float32)>]
478 6. `tf.RaggedTensor`: This is a composite tensor thats representation consists
479 of a flattened list of 'values' and a list of 'row_splits' which indicate how
480 to chop up the flattened list into different rows. For more details on
481 `tf.RaggedTensor`, please visit
482 https://www.tensorflow.org/api_docs/python/tf/RaggedTensor.
484 With `expand_composites=False`, we treat RaggedTensor as a scalar.
486 >>> structure = { "foo": tf.ragged.constant([[1, 2], [3]]),
487 ... "bar": tf.constant([[5]]) }
488 >>> flat_sequence = [ "one", "two" ]
489 >>> tf.nest.pack_sequence_as(structure, flat_sequence,
490 ... expand_composites=False)
491 {'foo': 'two', 'bar': 'one'}
493 With `expand_composites=True`, we expect that the flattened input contains
494 the tensors making up the ragged tensor i.e. the values and row_splits
495 tensors.
497 >>> structure = { "foo": tf.ragged.constant([[1., 2.], [3.]]),
498 ... "bar": tf.constant([[5.]]) }
499 >>> tensors = tf.nest.flatten(structure, expand_composites=True)
500 >>> print(tensors)
501 [<tf.Tensor: shape=(1, 1), dtype=float32, numpy=array([[5.]],
502 dtype=float32)>,
503 <tf.Tensor: shape=(3,), dtype=float32, numpy=array([1., 2., 3.],
504 dtype=float32)>,
505 <tf.Tensor: shape=(3,), dtype=int64, numpy=array([0, 2, 3])>]
506 >>> verified_tensors = [tf.debugging.check_numerics(t, 'invalid tensor: ')
507 ... if t.dtype==tf.float32 else t
508 ... for t in tensors]
509 >>> tf.nest.pack_sequence_as(structure, verified_tensors,
510 ... expand_composites=True)
511 {'foo': <tf.RaggedTensor [[1.0, 2.0], [3.0]]>,
512 'bar': <tf.Tensor: shape=(1, 1), dtype=float32, numpy=array([[5.]],
513 dtype=float32)>}
515 Args:
516 structure: Nested structure, whose structure is given by nested lists,
517 tuples, and dicts. Note: numpy arrays and strings are considered
518 scalars.
519 flat_sequence: flat sequence to pack.
520 expand_composites: If true, then composite tensors such as
521 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
522 component tensors.
524 Returns:
525 packed: `flat_sequence` converted to have the same recursive structure as
526 `structure`.
528 Raises:
529 ValueError: If `flat_sequence` and `structure` have different
530 atom counts.
531 TypeError: `structure` is or contains a dict with non-sortable keys.
532 """
533 return nest_util.pack_sequence_as(
534 nest_util.Modality.CORE, structure, flat_sequence, expand_composites
535 )
538@tf_export("nest.map_structure")
539def map_structure(func, *structure, **kwargs):
540 """Creates a new structure by applying `func` to each atom in `structure`.
542 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
543 for the definition of a structure.
545 Applies `func(x[0], x[1], ...)` where x[i] enumerates all atoms in
546 `structure[i]`. All items in `structure` must have the same arity,
547 and the return value will contain results with the same structure layout.
549 Examples:
551 * A single Python dict:
553 >>> a = {"hello": 24, "world": 76}
554 >>> tf.nest.map_structure(lambda p: p * 2, a)
555 {'hello': 48, 'world': 152}
557 * Multiple Python dictionaries:
559 >>> d1 = {"hello": 24, "world": 76}
560 >>> d2 = {"hello": 36, "world": 14}
561 >>> tf.nest.map_structure(lambda p1, p2: p1 + p2, d1, d2)
562 {'hello': 60, 'world': 90}
564 * A single Python list:
566 >>> a = [24, 76, "ab"]
567 >>> tf.nest.map_structure(lambda p: p * 2, a)
568 [48, 152, 'abab']
570 * Scalars:
572 >>> tf.nest.map_structure(lambda x, y: x + y, 3, 4)
573 7
575 * Empty structures:
577 >>> tf.nest.map_structure(lambda x: x + 1, ())
578 ()
580 * Check the types of iterables:
582 >>> s1 = (((1, 2), 3), 4, (5, 6))
583 >>> s1_list = [[[1, 2], 3], 4, [5, 6]]
584 >>> tf.nest.map_structure(lambda x, y: None, s1, s1_list)
585 Traceback (most recent call last):
586 ...
587 TypeError: The two structures don't have the same nested structure
589 * Type check is set to False:
591 >>> s1 = (((1, 2), 3), 4, (5, 6))
592 >>> s1_list = [[[1, 2], 3], 4, [5, 6]]
593 >>> tf.nest.map_structure(lambda x, y: None, s1, s1_list, check_types=False)
594 (((None, None), None), None, (None, None))
596 Args:
597 func: A callable that accepts as many arguments as there are structures.
598 *structure: atom or nested structure.
599 **kwargs: Valid keyword args are:
600 * `check_types`: If set to `True` (default) the types of iterables within
601 the structures have to be same (e.g. `map_structure(func, [1], (1,))`
602 raises a `TypeError` exception). To allow this set this argument to
603 `False`. Note that namedtuples with identical name and fields are always
604 considered to have the same shallow structure.
605 * `expand_composites`: If set to `True`, then composite tensors such as
606 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
607 component tensors. If `False` (the default), then composite tensors are
608 not expanded.
610 Returns:
611 A new structure with the same arity as `structure[0]`, whose atoms
612 correspond to `func(x[0], x[1], ...)` where `x[i]` is the atom in the
613 corresponding location in `structure[i]`. If there are different structure
614 types and `check_types` is `False` the structure types of the first
615 structure will be used.
617 Raises:
618 TypeError: If `func` is not callable or if the structures do not match
619 each other by depth tree.
620 ValueError: If no structure is provided or if the structures do not match
621 each other by type.
622 ValueError: If wrong keyword arguments are provided.
623 """
624 return nest_util.map_structure(
625 nest_util.Modality.CORE, func, *structure, **kwargs
626 )
629def map_structure_with_paths(func, *structure, **kwargs):
630 """Applies `func` to each entry in `structure` and returns a new structure.
632 Applies `func(path, x[0], x[1], ..., **kwargs)` where x[i] is an entry in
633 `structure[i]` and `path` is the common path to x[i] in the structures. All
634 structures in `structure` must have the same arity, and the return value will
635 contain the results with the same structure layout. Special kwarg
636 `check_types` determines whether the types of iterables within the structure
637 must be the same-- see **kwargs definition below.
639 Args:
640 func: A callable with the signature func(path, *values, **kwargs) that is
641 evaluated on the leaves of the structure.
642 *structure: A variable number of compatible structures to process.
643 **kwargs: Optional kwargs to be passed through to func. Special kwarg
644 `check_types` is not passed to func, but instead determines whether the
645 types of iterables within the structures have to be same (e.g.,
646 `map_structure(func, [1], (1,))` raises a `TypeError` exception). By
647 default, the types must match. To allow iteration over structures of
648 different types (but common arity), set this kwarg to `False`.
650 Returns:
651 A structure of the same form as the input structures whose leaves are the
652 result of evaluating func on corresponding leaves of the input structures.
654 Raises:
655 TypeError: If `func` is not callable or if the structures do not match
656 each other by depth tree.
657 TypeError: If `check_types` is not `False` and the two structures differ in
658 the type of sequence in any of their substructures.
659 ValueError: If no structures are provided.
660 """
661 def wrapper_func(tuple_path, *inputs, **kwargs):
662 string_path = "/".join(str(s) for s in tuple_path)
663 return func(string_path, *inputs, **kwargs)
665 return nest_util.map_structure_up_to(
666 nest_util.Modality.CORE, structure[0], wrapper_func, *structure, **kwargs
667 )
670def map_structure_with_tuple_paths(func, *structure, **kwargs):
671 """Applies `func` to each entry in `structure` and returns a new structure.
673 Applies `func(tuple_path, x[0], x[1], ..., **kwargs)` where `x[i]` is an entry
674 in `structure[i]` and `tuple_path` is a tuple of indices and/or dictionary
675 keys (as returned by `nest.yield_flat_paths`), which uniquely specifies the
676 common path to x[i] in the structures. All structures in `structure` must have
677 the same arity, and the return value will contain the results in the same
678 structure. Special kwarg `check_types` determines whether the types of
679 iterables within the structure must be the same-- see **kwargs definition
680 below.
682 Args:
683 func: A callable with the signature `func(tuple_path, *values, **kwargs)`
684 that is evaluated on the leaves of the structure.
685 *structure: A variable number of compatible structures to process.
686 **kwargs: Optional kwargs to be passed through to func. Special kwarg
687 `check_types` is not passed to func, but instead determines whether the
688 types of iterables within the structures have to be same (e.g.
689 `map_structure(func, [1], (1,))` raises a `TypeError` exception). To allow
690 this set this argument to `False`.
692 Returns:
693 A structure of the same form as the input structures whose leaves are the
694 result of evaluating func on corresponding leaves of the input structures.
696 Raises:
697 TypeError: If `func` is not callable or if the structures do not match
698 each other by depth tree.
699 TypeError: If `check_types` is not `False` and the two structures differ in
700 the type of sequence in any of their substructures.
701 ValueError: If no structures are provided.
702 """
703 return nest_util.map_structure_up_to(
704 nest_util.Modality.CORE, structure[0], func, *structure, **kwargs
705 )
708def assert_shallow_structure(shallow_tree,
709 input_tree,
710 check_types=True,
711 expand_composites=False):
712 """Asserts that `shallow_tree` is a shallow structure of `input_tree`.
714 That is, this function tests if the `input_tree` structure can be created from
715 the `shallow_tree` structure by replacing its leaf nodes with deeper
716 tree structures.
718 Examples:
720 The following code will raise an exception:
721 ```python
722 shallow_tree = {"a": "A", "b": "B"}
723 input_tree = {"a": 1, "c": 2}
724 assert_shallow_structure(shallow_tree, input_tree)
725 ```
727 The following code will raise an exception:
728 ```python
729 shallow_tree = ["a", "b"]
730 input_tree = ["c", ["d", "e"], "f"]
731 assert_shallow_structure(shallow_tree, input_tree)
732 ```
734 Args:
735 shallow_tree: an arbitrarily nested structure.
736 input_tree: an arbitrarily nested structure.
737 check_types: if `True` (default) the sequence types of `shallow_tree` and
738 `input_tree` have to be the same. Note that even with check_types==True,
739 this function will consider two different namedtuple classes with the same
740 name and _fields attribute to be the same class.
741 expand_composites: If true, then composite tensors such as
742 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
743 component tensors.
744 Raises:
745 TypeError: If `shallow_tree` is a sequence but `input_tree` is not.
746 TypeError: If the sequence types of `shallow_tree` are different from
747 `input_tree`. Only raised if `check_types` is `True`.
748 ValueError: If the sequence lengths of `shallow_tree` are different from
749 `input_tree`.
750 """
751 nest_util.assert_shallow_structure(
752 nest_util.Modality.CORE,
753 shallow_tree,
754 input_tree,
755 check_types,
756 expand_composites,
757 )
760@tf_export("__internal__.nest.flatten_up_to", v1=[])
761def flatten_up_to(shallow_tree, input_tree, check_types=True,
762 expand_composites=False):
763 """Flattens `input_tree` up to `shallow_tree`.
765 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
766 for the definition of a structure.
768 Any further depth in structure in `input_tree` is retained as structures in
769 the partially flatten output.
771 If `shallow_tree` and `input_tree` are atoms, this returns a
772 single-item list: `[input_tree]`.
774 Use Case:
776 Sometimes we may wish to partially flatten a structure, retaining some
777 of the nested structure. We achieve this by specifying a shallow structure,
778 `shallow_tree`, we wish to flatten up to.
780 The input, `input_tree`, can be thought of as having the same structure layout
781 as `shallow_tree`, but with leaf nodes that are themselves tree structures.
783 Examples:
785 ```python
786 input_tree = [[[2, 2], [3, 3]], [[4, 9], [5, 5]]]
787 shallow_tree = [[True, True], [False, True]]
789 flattened_input_tree = flatten_up_to(shallow_tree, input_tree)
790 flattened_shallow_tree = flatten_up_to(shallow_tree, shallow_tree)
792 # Output is:
793 # [[2, 2], [3, 3], [4, 9], [5, 5]]
794 # [True, True, False, True]
795 ```
797 ```python
798 input_tree = [[('a', 1), [('b', 2), [('c', 3), [('d', 4)]]]]]
799 shallow_tree = [['level_1', ['level_2', ['level_3', ['level_4']]]]]
801 input_tree_flattened_as_shallow_tree = flatten_up_to(shallow_tree, input_tree)
802 input_tree_flattened = flatten(input_tree)
804 # Output is:
805 # [('a', 1), ('b', 2), ('c', 3), ('d', 4)]
806 # ['a', 1, 'b', 2, 'c', 3, 'd', 4]
807 ```
809 Edge Cases for atoms:
811 ```python
812 flatten_up_to(0, 0) # Output: [0]
813 flatten_up_to(0, [0, 1, 2]) # Output: [[0, 1, 2]]
814 flatten_up_to([0, 1, 2], 0) # Output: TypeError
815 flatten_up_to([0, 1, 2], [0, 1, 2]) # Output: [0, 1, 2]
816 ```
818 Args:
819 shallow_tree: a possibly pruned structure of input_tree.
820 input_tree: an atom or a nested structure.
821 Note, numpy arrays are considered atoms.
822 check_types: bool. If True, check that each node in shallow_tree has the
823 same type as the corresponding node in input_tree.
824 expand_composites: If true, then composite tensors such as
825 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
826 component tensors.
828 Returns:
829 A Python list, the partially flattened version of `input_tree` according to
830 the structure of `shallow_tree`.
832 Raises:
833 TypeError: If `shallow_tree` is a nested structure but `input_tree` is not.
834 TypeError: If the structure types of `shallow_tree` are different from
835 `input_tree`.
836 ValueError: If the structure lengths of `shallow_tree` are different from
837 `input_tree`.
838 """
839 return nest_util.flatten_up_to(
840 nest_util.Modality.CORE,
841 shallow_tree,
842 input_tree,
843 check_types,
844 expand_composites,
845 )
848def flatten_with_tuple_paths_up_to(shallow_tree,
849 input_tree,
850 check_types=True,
851 expand_composites=False):
852 """Flattens `input_tree` up to `shallow_tree`.
854 Any further depth in structure in `input_tree` is retained as structures in
855 the partially flattened output.
857 Returns a list of (path, value) pairs, where value a leaf node in the
858 flattened tree, and path is the tuple path of that leaf in input_tree.
860 If `shallow_tree` and `input_tree` are not sequences, this returns a
861 single-item list: `[((), input_tree)]`.
863 Use Case:
865 Sometimes we may wish to partially flatten a nested sequence, retaining some
866 of the nested structure. We achieve this by specifying a shallow structure,
867 `shallow_tree`, we wish to flatten up to.
869 The input, `input_tree`, can be thought of as having the same structure layout
870 as `shallow_tree`, but with leaf nodes that are themselves tree structures.
872 Examples:
874 ```python
875 input_tree = [[[2, 2], [3, 3]], [[4, 9], [5, 5]]]
876 shallow_tree = [[True, True], [False, True]]
878 flattened_input_tree = flatten_with_tuple_paths_up_to(shallow_tree,
879 input_tree)
880 flattened_shallow_tree = flatten_with_tuple_paths_up_to(shallow_tree,
881 shallow_tree)
883 # Output is:
884 # [((0, 0), [2, 2]),
885 # ((0, 1), [3, 3]),
886 # ((1, 0), [4, 9]),
887 # ((1, 1), [5, 5])]
888 #
889 # [((0, 0), True),
890 # ((0, 1), True),
891 # ((1, 0), False),
892 # ((1, 1), True)]
893 ```
895 ```python
896 input_tree = [[('a', 1), [('b', 2), [('c', 3), [('d', 4)]]]]]
897 shallow_tree = [['level_1', ['level_2', ['level_3', ['level_4']]]]]
899 input_tree_flattened_as_shallow_tree = flatten_up_to(shallow_tree, input_tree)
900 input_tree_flattened = flatten(input_tree)
902 # Output is:
903 # [((0, 0), ('a', 1)),
904 # ((0, 1, 0), ('b', 2)),
905 # ((0, 1, 1, 0), ('c', 3)),
906 # ((0, 1, 1, 1), ('d', 4))]
907 # ['a', 1, 'b', 2, 'c', 3, 'd', 4]
908 ```
910 Non-Sequence Edge Cases:
912 ```python
913 flatten_with_tuple_paths_up_to(0, 0) # Output: [(), 0]
915 flatten_with_tuple_paths_up_to(0, [0, 1, 2]) # Output: [(), [0, 1, 2]]
917 flatten_with_tuple_paths_up_to([0, 1, 2], 0) # Output: TypeError
919 flatten_with_tuple_paths_up_to([0, 1, 2], [0, 1, 2])
920 # Output: [((0,) 0), ((1,), 1), ((2,), 2)]
921 ```
923 Args:
924 shallow_tree: a possibly pruned structure of input_tree.
925 input_tree: an atom or a nested structure.
926 Note, numpy arrays are considered atoms.
927 check_types: bool. If True, check that each node in shallow_tree has the
928 same type as the corresponding node in input_tree.
929 expand_composites: If true, then composite tensors such as
930 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
931 component tensors.
933 Returns:
934 A Python list, the partially flattened version of `input_tree` according to
935 the structure of `shallow_tree`.
937 Raises:
938 TypeError: If `shallow_tree` is a nested structure but `input_tree` is not.
939 TypeError: If the structure types of `shallow_tree` are different from
940 `input_tree`.
941 ValueError: If the structure lengths of `shallow_tree` are different from
942 `input_tree`.
943 """
944 is_nested_fn = _is_nested_or_composite if expand_composites else is_nested
945 assert_shallow_structure(shallow_tree,
946 input_tree,
947 check_types=check_types,
948 expand_composites=expand_composites)
949 return list(
950 nest_util.yield_flat_up_to(
951 nest_util.Modality.CORE, shallow_tree, input_tree, is_nested_fn
952 )
953 )
956@tf_export("__internal__.nest.map_structure_up_to", v1=[])
957def map_structure_up_to(shallow_tree, func, *inputs, **kwargs):
958 """Applies a function or op to a number of partially flattened inputs.
960 The `inputs` are flattened up to `shallow_tree` before being mapped.
962 Use Case:
964 Sometimes we wish to apply a function to a partially flattened
965 structure (for example when the function itself takes structure inputs). We
966 achieve this by specifying a shallow structure, `shallow_tree` we wish to
967 flatten up to.
969 The `inputs`, can be thought of as having the same structure layout as
970 `shallow_tree`, but with leaf nodes that are themselves tree structures.
972 This function therefore will return something with the same base structure as
973 `shallow_tree`.
975 Examples:
977 ```python
978 shallow_tree = [None, None]
979 inp_val = [1, 2, 3]
980 out = map_structure_up_to(shallow_tree, lambda x: 2 * x, inp_val)
982 # Output is: [2, 4]
983 ```
985 ```python
986 ab_tuple = collections.namedtuple("ab_tuple", "a, b")
987 op_tuple = collections.namedtuple("op_tuple", "add, mul")
988 inp_val = ab_tuple(a=2, b=3)
989 inp_ops = ab_tuple(a=op_tuple(add=1, mul=2), b=op_tuple(add=2, mul=3))
990 out = map_structure_up_to(inp_val, lambda val, ops: (val + ops.add) * ops.mul,
991 inp_val, inp_ops)
993 # Output is: ab_tuple(a=6, b=15)
994 ```
996 ```python
997 data_list = [[2, 4, 6, 8], [[1, 3, 5, 7, 9], [3, 5, 7]]]
998 name_list = ['evens', ['odds', 'primes']]
999 out = map_structure_up_to(
1000 name_list,
1001 lambda name, sec: "first_{}_{}".format(len(sec), name),
1002 name_list, data_list)
1004 # Output is: ['first_4_evens', ['first_5_odds', 'first_3_primes']]
1005 ```
1007 Args:
1008 shallow_tree: a shallow structure, common to all the inputs.
1009 func: callable which will be applied to each input individually.
1010 *inputs: structures that are compatible with shallow_tree. The function
1011 `func` is applied to corresponding structures due to partial flattening
1012 of each input, so the function must support arity of `len(inputs)`.
1013 **kwargs: kwargs to feed to func(). Special kwarg
1014 `check_types` is not passed to func, but instead determines whether the
1015 types of iterables within the structures have to be same (e.g.
1016 `map_structure(func, [1], (1,))` raises a `TypeError` exception). To allow
1017 this set this argument to `False`.
1019 Raises:
1020 TypeError: If `shallow_tree` is a nested structure but `input_tree` is not.
1021 TypeError: If the structure types of `shallow_tree` are different from
1022 `input_tree`.
1023 ValueError: If the structure lengths of `shallow_tree` are different from
1024 `input_tree`.
1026 Returns:
1027 result of repeatedly applying `func`, with the same structure layout as
1028 `shallow_tree`.
1029 """
1030 return nest_util.map_structure_up_to(
1031 nest_util.Modality.CORE,
1032 shallow_tree,
1033 lambda _, *values: func(*values), # Discards the path arg.
1034 *inputs,
1035 **kwargs,
1036 )
1039def map_structure_with_tuple_paths_up_to(shallow_tree, func, *inputs, **kwargs):
1040 """Applies a function or op to a number of partially flattened inputs.
1042 Like map_structure_up_to(), except that the 'func' argument takes a path
1043 tuple as its first argument, followed by the corresponding values from
1044 *inputs.
1046 Example:
1048 ```python
1049 lowercase = {'a': 'a', 'b': ('b0', 'b1')}
1050 uppercase = {'a': 'A', 'b': ('B0', 'B1')}
1052 def print_path_and_values(path, *values):
1053 print("path: {}, values: {}".format(path, values))
1055 shallow_tree = {'a': None}
1056 map_structure_with_tuple_paths_up_to(shallow_tree,
1057 print_path_and_values,
1058 lowercase,
1059 uppercase)
1060 path: ('a',), values: ('a', 'A')
1061 path: ('b', 0), values: ('b0', 'B0')
1062 path: ('b', 1), values: ('b1', 'B1')
1064 shallow_tree = {'b': None}
1065 map_structure_with_tuple_paths_up_to(shallow_tree,
1066 print_path_and_values,
1067 lowercase,
1068 uppercase,
1069 check_types=False)
1070 path: ('b', 1), values: (('bo', 'b1'), ('B0', 'B1'))
1072 shallow_tree = {'a': None, 'b': {1: None}}
1073 map_structure_with_tuple_paths_up_to(shallow_tree,
1074 print_path_and_values,
1075 lowercase,
1076 uppercase,
1077 check_types=False)
1078 path: ('a',), values: ('a', 'A')
1079 path: ('b', 1), values: ('b1', B1')
1080 ```
1082 Args:
1083 shallow_tree: a shallow structure, common to all the inputs.
1084 func: callable that takes args (path, inputs_0_value, ... , inputs_N_value),
1085 where path is a tuple path to an atom in shallow_tree, and inputs_i_value
1086 is the corresponding value from inputs[i].
1087 *inputs: structures that are all structurally compatible with shallow_tree.
1088 **kwargs: kwargs to feed to func(). Special kwarg `check_types` is not
1089 passed to func, but instead determines whether the types of iterables
1090 within the structures have to be same (e.g. `map_structure(func, [1],
1091 (1,))` raises a `TypeError` exception). To allow this set this argument to
1092 `False`.
1094 Raises:
1095 TypeError: If `shallow_tree` is a nested structure but one of `*inputs` is
1096 not.
1097 TypeError: If the structure types of `shallow_tree` are different from
1098 `input_tree`.
1099 ValueError: If the structure lengths of `shallow_tree` are different from
1100 `input_tree`.
1102 Returns:
1103 Result of repeatedly applying `func`. Has the same structure layout as
1104 `shallow_tree`.
1105 """
1106 return nest_util.map_structure_up_to(
1107 nest_util.Modality.CORE, shallow_tree, func, *inputs, **kwargs
1108 )
1111@tf_export("__internal__.nest.get_traverse_shallow_structure", v1=[])
1112def get_traverse_shallow_structure(traverse_fn, structure,
1113 expand_composites=False):
1114 """Generates a shallow structure from a `traverse_fn` and `structure`.
1116 `traverse_fn` must accept any possible subtree of `structure` and return
1117 a depth=1 structure containing `True` or `False` values, describing which
1118 of the top-level subtrees may be traversed. It may also
1119 return scalar `True` or `False` "traversal is OK / not OK for all subtrees."
1121 Examples are available in the unit tests (nest_test.py).
1123 Args:
1124 traverse_fn: Function taking a substructure and returning either a scalar
1125 `bool` (whether to traverse that substructure or not) or a depth=1
1126 shallow structure of the same type, describing which parts of the
1127 substructure to traverse.
1128 structure: The structure to traverse.
1129 expand_composites: If true, then composite tensors such as
1130 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
1131 component tensors.
1133 Returns:
1134 A shallow structure containing python bools, which can be passed to
1135 `map_structure_up_to` and `flatten_up_to`.
1137 Raises:
1138 TypeError: if `traverse_fn` returns a nested structure for an atom input.
1139 or a structure with depth higher than 1 for a nested structure input,
1140 or if any leaf values in the returned structure or scalar are not type
1141 `bool`.
1142 """
1143 is_nested_fn = _is_nested_or_composite if expand_composites else is_nested
1144 to_traverse = traverse_fn(structure)
1145 if not is_nested_fn(structure):
1146 if not isinstance(to_traverse, bool):
1147 raise TypeError("traverse_fn returned structure: %s for non-structure: %s"
1148 % (to_traverse, structure))
1149 return to_traverse
1150 level_traverse = []
1151 if isinstance(to_traverse, bool):
1152 if not to_traverse:
1153 # Do not traverse this substructure at all. Exit early.
1154 return False
1155 else:
1156 # Traverse the entire substructure.
1157 for branch in nest_util.yield_value(nest_util.Modality.CORE, structure):
1158 level_traverse.append(
1159 get_traverse_shallow_structure(traverse_fn, branch,
1160 expand_composites=expand_composites))
1161 elif not is_nested_fn(to_traverse):
1162 raise TypeError("traverse_fn returned a non-bool scalar: %s for input: %s"
1163 % (to_traverse, structure))
1164 else:
1165 # Traverse some subset of this substructure.
1166 assert_shallow_structure(to_traverse, structure,
1167 expand_composites=expand_composites)
1168 for t, branch in zip(
1169 nest_util.yield_value(nest_util.Modality.CORE, to_traverse),
1170 nest_util.yield_value(nest_util.Modality.CORE, structure),
1171 ):
1172 if not isinstance(t, bool):
1173 raise TypeError(
1174 "traverse_fn didn't return a depth=1 structure of bools. saw: %s "
1175 " for structure: %s" % (to_traverse, structure))
1176 if t:
1177 level_traverse.append(
1178 get_traverse_shallow_structure(traverse_fn, branch))
1179 else:
1180 level_traverse.append(False)
1181 return nest_util.sequence_like(structure, level_traverse)
1184@tf_export("__internal__.nest.yield_flat_paths", v1=[])
1185def yield_flat_paths(nest, expand_composites=False):
1186 """Yields paths for some nested structure.
1188 Refer to [tf.nest](https://www.tensorflow.org/api_docs/python/tf/nest)
1189 for the definition of a structure.
1191 Paths are lists of objects which can be str-converted, which may include
1192 integers or other types which are used as indices in a dict.
1194 The flat list will be in the corresponding order as if you called
1195 `nest.flatten` on the structure. This is handy for naming Tensors such
1196 the TF scope structure matches the tuple structure.
1198 E.g. if we have a tuple `value = Foo(a=3, b=Bar(c=23, d=42))`
1200 ```shell
1201 nest.flatten(value)
1202 [3, 23, 42]
1203 list(nest.yield_flat_paths(value))
1204 [('a',), ('b', 'c'), ('b', 'd')]
1205 ```
1207 ```shell
1208 list(nest.yield_flat_paths({'a': [3]}))
1209 [('a', 0)]
1210 list(nest.yield_flat_paths({'a': 3}))
1211 [('a',)]
1212 ```
1214 Args:
1215 nest: the value to produce a flattened paths list for.
1216 expand_composites: If true, then composite tensors such as
1217 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
1218 component tensors.
1220 Yields:
1221 Tuples containing index or key values which form the path to a specific
1222 leaf value in the nested structure.
1223 """
1224 is_nested_fn = _is_nested_or_composite if expand_composites else is_nested
1225 for k, _ in nest_util.yield_flat_up_to(
1226 nest_util.Modality.CORE, nest, nest, is_nested_fn
1227 ):
1228 yield k
1231def flatten_with_joined_string_paths(structure, separator="/",
1232 expand_composites=False):
1233 """Returns a list of (string path, atom) tuples.
1235 The order of tuples produced matches that of `nest.flatten`. This allows you
1236 to flatten a nested structure while keeping information about where in the
1237 structure each atom was located. See `nest.yield_flat_paths`
1238 for more information.
1240 Args:
1241 structure: the nested structure to flatten.
1242 separator: string to separate levels of hierarchy in the results, defaults
1243 to '/'.
1244 expand_composites: If true, then composite tensors such as
1245 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
1246 component tensors.
1248 Returns:
1249 A list of (string, atom) tuples.
1250 """
1251 flat_paths = yield_flat_paths(structure, expand_composites=expand_composites)
1252 def stringify_and_join(path_elements):
1253 return separator.join(str(path_element) for path_element in path_elements)
1255 flat_string_paths = (stringify_and_join(path) for path in flat_paths)
1256 return list(zip(flat_string_paths,
1257 flatten(structure, expand_composites=expand_composites)))
1260def flatten_with_tuple_paths(structure, expand_composites=False):
1261 """Returns a list of `(tuple_path, atom)` tuples.
1263 The order of pairs produced matches that of `nest.flatten`. This allows you
1264 to flatten a nested structure while keeping information about where in the
1265 structure each atom was located. See `nest.yield_flat_paths`
1266 for more information about tuple paths.
1268 Args:
1269 structure: the nested structure to flatten.
1270 expand_composites: If true, then composite tensors such as
1271 `tf.sparse.SparseTensor` and `tf.RaggedTensor` are expanded into their
1272 component tensors.
1274 Returns:
1275 A list of `(tuple_path, atom)` tuples. Each `tuple_path` is a tuple
1276 of indices and/or dictionary keys that uniquely specify the path to
1277 `atom` within `structure`.
1278 """
1279 return list(zip(yield_flat_paths(structure,
1280 expand_composites=expand_composites),
1281 flatten(structure, expand_composites=expand_composites)))
1284@tf_export("__internal__.nest.list_to_tuple", v1=[])
1285def list_to_tuple(structure):
1286 """Replace all lists with tuples.
1288 The fork of nest that tf.data uses treats lists as atoms, while
1289 tf.nest treats them as structures to recurse into. Keras has chosen to adopt
1290 the latter convention, and must therefore deeply replace all lists with tuples
1291 before passing structures to Dataset.from_generator.
1293 Args:
1294 structure: A nested structure to be remapped.
1296 Returns:
1297 structure mapped to replace all lists with tuples.
1298 """
1299 def sequence_fn(instance, args):
1300 if isinstance(instance, list):
1301 return tuple(args)
1302 return nest_util.sequence_like(instance, args)
1304 return nest_util.pack_sequence_as(
1305 nest_util.Modality.CORE,
1306 structure,
1307 flatten(structure),
1308 False,
1309 sequence_fn=sequence_fn,
1310 )
1313_pywrap_utils.RegisterType("Mapping", _collections_abc.Mapping)
1314_pywrap_utils.RegisterType("MutableMapping", _collections_abc.MutableMapping)
1315_pywrap_utils.RegisterType("Sequence", _collections_abc.Sequence)
1316_pywrap_utils.RegisterType("MappingView", _collections_abc.MappingView)
1317_pywrap_utils.RegisterType("ObjectProxy", _wrapt.ObjectProxy)