Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/tensorflow/python/ops/ragged/ragged_operators.py: 84%
58 statements
« prev ^ index » next coverage.py v7.4.0, created at 2024-01-03 07:57 +0000
1# Copyright 2018 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# ==============================================================================
15"""Operator overloads for `RaggedTensor`."""
17from tensorflow.python.framework import ops
18from tensorflow.python.ops import math_ops
19from tensorflow.python.ops.ragged import ragged_getitem
20from tensorflow.python.ops.ragged import ragged_tensor
21from tensorflow.python.util import tf_decorator
24# =============================================================================
25# Equality Docstring
26# =============================================================================
27def ragged_eq(self, other): # pylint: disable=g-doc-args
28 """Returns result of elementwise `==` or False if not broadcast-compatible.
30 Compares two ragged tensors elemewise for equality if they are
31 broadcast-compatible; or returns False if they are not
32 [broadcast-compatible](http://docs.scipy.org/doc/numpy/user/basics.broadcasting.html).
34 Note that this behavior differs from `tf.math.equal`, which raises an
35 exception if the two ragged tensors are not broadcast-compatible.
37 For example:
39 >>> rt1 = tf.ragged.constant([[1, 2], [3]])
40 >>> rt1 == rt1
41 <tf.RaggedTensor [[True, True], [True]]>
43 >>> rt2 = tf.ragged.constant([[1, 2], [4]])
44 >>> rt1 == rt2
45 <tf.RaggedTensor [[True, True], [False]]>
47 >>> rt3 = tf.ragged.constant([[1, 2], [3, 4]])
48 >>> # rt1 and rt3 are not broadcast-compatible.
49 >>> rt1 == rt3
50 False
52 >>> # You can also compare a `tf.RaggedTensor` to a `tf.Tensor`.
53 >>> t = tf.constant([[1, 2], [3, 4]])
54 >>> rt1 == t
55 False
56 >>> t == rt1
57 False
58 >>> rt4 = tf.ragged.constant([[1, 2], [3, 4]])
59 >>> rt4 == t
60 <tf.RaggedTensor [[True, True], [True, True]]>
61 >>> t == rt4
62 <tf.RaggedTensor [[True, True], [True, True]]>
64 Args:
65 other: The right-hand side of the `==` operator.
67 Returns:
68 The ragged tensor result of the elementwise `==` operation, or `False` if
69 the arguments are not broadcast-compatible.
70 """
71 return math_ops.tensor_equals(self, other)
74# =============================================================================
75# Ordering Docstring
76# =============================================================================
77def ragged_ge(self, other): # pylint: disable=g-doc-args
78 """Elementwise `>=` comparison of two convertible-to-ragged-tensor values.
80 Computes the elemewise `>=` comparison of two values that are convertible to
81 ragged tenors, with [broadcasting]
82 (http://docs.scipy.org/doc/numpy/user/basics.broadcasting.html) support.
83 Raises an exception if two values are not broadcast-compatible.
85 For example:
87 >>> rt1 = tf.ragged.constant([[1, 2], [3]])
88 >>> rt1 >= rt1
89 <tf.RaggedTensor [[True, True], [True]]>
91 >>> rt2 = tf.ragged.constant([[2, 1], [3]])
92 >>> rt1 >= rt2
93 <tf.RaggedTensor [[False, True], [True]]>
95 >>> rt3 = tf.ragged.constant([[1, 2], [3, 4]])
96 >>> # rt1 and rt3 are not broadcast-compatible.
97 >>> rt1 >= rt3
98 Traceback (most recent call last):
99 ...
100 InvalidArgumentError: ...
102 >>> # You can also compare a `tf.RaggedTensor` to a `tf.Tensor`.
103 >>> rt4 = tf.ragged.constant([[1, 2],[3, 4]])
104 >>> t1 = tf.constant([[2, 1], [4, 3]])
105 >>> rt4 >= t1
106 <tf.RaggedTensor [[False, True],
107 [False, True]]>
108 >>> t1 >= rt4
109 <tf.RaggedTensor [[True, False],
110 [True, False]]>
112 >>> # Compares a `tf.RaggedTensor` to a `tf.Tensor` with broadcasting.
113 >>> t2 = tf.constant([[2]])
114 >>> rt4 >= t2
115 <tf.RaggedTensor [[False, True],
116 [True, True]]>
117 >>> t2 >= rt4
118 <tf.RaggedTensor [[True, True],
119 [False, False]]>
121 Args:
122 other: The right-hand side of the `>=` operator.
124 Returns:
125 A `tf.RaggedTensor` of dtype `tf.bool` with the shape that `self` and
126 `other` broadcast to.
128 Raises:
129 InvalidArgumentError: If `self` and `other` are not broadcast-compatible.
130 """
131 return math_ops.greater_equal(self, other)
134# =============================================================================
135# Logical Docstring
136# =============================================================================
139# =============================================================================
140# Arithmetic Docstring
141# =============================================================================
142def ragged_abs(self, name=None): # pylint: disable=g-doc-args
143 r"""Computes the absolute value of a ragged tensor.
145 Given a ragged tensor of integer or floating-point values, this operation
146 returns a ragged tensor of the same type, where each element contains the
147 absolute value of the corresponding element in the input.
149 Given a ragged tensor `x` of complex numbers, this operation returns a tensor
150 of type `float32` or `float64` that is the absolute value of each element in
151 `x`. For a complex number \\(a + bj\\), its absolute value is computed as
152 \\(\sqrt{a^2 + b^2}\\).
154 For example:
156 >>> # real number
157 >>> x = tf.ragged.constant([[-2.2, 3.2], [-4.2]])
158 >>> tf.abs(x)
159 <tf.RaggedTensor [[2.2, 3.2], [4.2]]>
161 >>> # complex number
162 >>> x = tf.ragged.constant([[-2.2 + 4.7j], [-3.2 + 5.7j], [-4.2 + 6.7j]])
163 >>> tf.abs(x)
164 <tf.RaggedTensor [[5.189412298131649],
165 [6.536818798161687],
166 [7.907591289387685]]>
168 Args:
169 name: A name for the operation (optional).
171 Returns:
172 A `RaggedTensor` of the same size and type as `x`, with absolute values.
173 Note, for `complex64` or `complex128` input, the returned `RaggedTensor`
174 will be of type `float32` or `float64`, respectively.
175 """
176 return math_ops.abs(self, name=name)
179# ===========================================================================
180def ragged_and(self, y, name=None): # pylint: disable=g-doc-args
181 r"""Returns the truth value of elementwise `x & y`.
183 Logical AND function.
185 Requires that `x` and `y` have the same shape or have
186 [broadcast-compatible](http://docs.scipy.org/doc/numpy/user/basics.broadcasting.html)
187 shapes. For example, `y` can be:
189 - A single Python boolean, where the result will be calculated by applying
190 logical AND with the single element to each element in `x`.
191 - A `tf.Tensor` object of dtype `tf.bool` of the same shape or
192 [broadcast-compatible](http://docs.scipy.org/doc/numpy/user/basics.broadcasting.html)
193 shape. In this case, the result will be the element-wise logical AND of
194 `x` and `y`.
195 - A `tf.RaggedTensor` object of dtype `tf.bool` of the same shape or
196 [broadcast-compatible](http://docs.scipy.org/doc/numpy/user/basics.broadcasting.html)
197 shape. In this case, the result will be the element-wise logical AND of
198 `x` and `y`.
200 For example:
202 >>> # `y` is a Python boolean
203 >>> x = tf.ragged.constant([[True, False], [True]])
204 >>> y = True
205 >>> x & y
206 <tf.RaggedTensor [[True, False], [True]]>
207 >>> tf.math.logical_and(x, y) # Equivalent of x & y
208 <tf.RaggedTensor [[True, False], [True]]>
209 >>> y & x
210 <tf.RaggedTensor [[True, False], [True]]>
211 >>> tf.math.reduce_all(x & y) # Reduce to a scalar bool Tensor.
212 <tf.Tensor: shape=(), dtype=bool, numpy=False>
214 >>> # `y` is a tf.Tensor of the same shape.
215 >>> x = tf.ragged.constant([[True, False], [True, False]])
216 >>> y = tf.constant([[True, False], [False, True]])
217 >>> x & y
218 <tf.RaggedTensor [[True, False], [False, False]]>
220 >>> # `y` is a tf.Tensor of a broadcast-compatible shape.
221 >>> x = tf.ragged.constant([[True, False], [True]])
222 >>> y = tf.constant([[True], [False]])
223 >>> x & y
224 <tf.RaggedTensor [[True, False], [False]]>
226 >>> # `y` is a `tf.RaggedTensor` of the same shape.
227 >>> x = tf.ragged.constant([[True, False], [True]])
228 >>> y = tf.ragged.constant([[False, True], [True]])
229 >>> x & y
230 <tf.RaggedTensor [[False, False], [True]]>
232 >>> # `y` is a `tf.RaggedTensor` of a broadcast-compatible shape.
233 >>> x = tf.ragged.constant([[[True, True, False]], [[]], [[True, False]]])
234 >>> y = tf.ragged.constant([[[True]], [[True]], [[False]]], ragged_rank=1)
235 >>> x & y
236 <tf.RaggedTensor [[[True, True, False]], [[]], [[False, False]]]>
238 Args:
239 y: A Python boolean or a `tf.Tensor` or `tf.RaggedTensor` of dtype
240 `tf.bool`.
241 name: A name for the operation (optional).
243 Returns:
244 A `tf.RaggedTensor` of dtype `tf.bool` with the shape that `x` and `y`
245 broadcast to.
246 """
247 return math_ops.logical_and(self, y, name)
250# Helper Methods.
251def _right(operator):
252 """Right-handed version of an operator: swap args x and y."""
253 return tf_decorator.make_decorator(operator, lambda y, x: operator(x, y))
256def ragged_hash(self):
257 """The operation invoked by the `RaggedTensor.__hash__` operator."""
258 g = getattr(self.row_splits, "graph", None)
259 # pylint: disable=protected-access
260 if (ops.Tensor._USE_EQUALITY and ops.executing_eagerly_outside_functions() and
261 (g is None or g.building_function)):
262 raise TypeError("RaggedTensor is unhashable.")
263 else:
264 return id(self)
267# Indexing
268ragged_tensor.RaggedTensor.__getitem__ = ragged_getitem.ragged_tensor_getitem
270# Equality
271ragged_tensor.RaggedTensor.__eq__ = ragged_eq
272ragged_tensor.RaggedTensor.__ne__ = math_ops.tensor_not_equals
273ragged_tensor.RaggedTensor.__hash__ = ragged_hash
275# Ordering operators
276ragged_tensor.RaggedTensor.__ge__ = ragged_ge
277ragged_tensor.RaggedTensor.__gt__ = math_ops.greater
278ragged_tensor.RaggedTensor.__le__ = math_ops.less_equal
279ragged_tensor.RaggedTensor.__lt__ = math_ops.less
281# Logical operators
282ragged_tensor.RaggedTensor.__and__ = ragged_and
283ragged_tensor.RaggedTensor.__rand__ = _right(ragged_and)
285ragged_tensor.RaggedTensor.__invert__ = math_ops.logical_not
286ragged_tensor.RaggedTensor.__ror__ = _right(math_ops.logical_or)
287ragged_tensor.RaggedTensor.__or__ = math_ops.logical_or
288ragged_tensor.RaggedTensor.__xor__ = math_ops.logical_xor
289ragged_tensor.RaggedTensor.__rxor__ = _right(math_ops.logical_xor)
291# Arithmetic operators
292ragged_tensor.RaggedTensor.__abs__ = ragged_abs
293ragged_tensor.RaggedTensor.__add__ = math_ops.add
294ragged_tensor.RaggedTensor.__radd__ = _right(math_ops.add)
295ragged_tensor.RaggedTensor.__div__ = math_ops.div
296ragged_tensor.RaggedTensor.__rdiv__ = _right(math_ops.div)
297ragged_tensor.RaggedTensor.__floordiv__ = math_ops.floordiv
298ragged_tensor.RaggedTensor.__rfloordiv__ = _right(math_ops.floordiv)
299ragged_tensor.RaggedTensor.__mod__ = math_ops.floormod
300ragged_tensor.RaggedTensor.__rmod__ = _right(math_ops.floormod)
301ragged_tensor.RaggedTensor.__mul__ = math_ops.multiply
302ragged_tensor.RaggedTensor.__rmul__ = _right(math_ops.multiply)
303ragged_tensor.RaggedTensor.__neg__ = math_ops.negative
304ragged_tensor.RaggedTensor.__pow__ = math_ops.pow
305ragged_tensor.RaggedTensor.__rpow__ = _right(math_ops.pow)
306ragged_tensor.RaggedTensor.__sub__ = math_ops.subtract
307ragged_tensor.RaggedTensor.__rsub__ = _right(math_ops.subtract)
308ragged_tensor.RaggedTensor.__truediv__ = math_ops.truediv
309ragged_tensor.RaggedTensor.__rtruediv__ = _right(math_ops.truediv)
312def ragged_bool(self): # pylint: disable=g-doc-args
313 """Raises TypeError when a RaggedTensor is used as a Python bool.
315 To prevent RaggedTensor from being used as a bool, this function always raise
316 TypeError when being called.
318 For example:
320 >>> x = tf.ragged.constant([[1, 2], [3]])
321 >>> result = True if x else False # Evaluate x as a bool value.
322 Traceback (most recent call last):
323 ...
324 TypeError: RaggedTensor may not be used as a boolean.
326 >>> x = tf.ragged.constant([[1]])
327 >>> r = (x == 1) # tf.RaggedTensor [[True]]
328 >>> if r: # Evaluate r as a bool value.
329 ... pass
330 Traceback (most recent call last):
331 ...
332 TypeError: RaggedTensor may not be used as a boolean.
333 """
334 raise TypeError("RaggedTensor may not be used as a boolean.")
337ragged_tensor.RaggedTensor.__bool__ = ragged_bool # Python3 bool conversion.
338ragged_tensor.RaggedTensor.__nonzero__ = ragged_bool # Python2 bool conversion.