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"""Private convenience functions for RaggedTensors. 16 17None of these methods are exposed in the main "ragged" package. 18""" 19 20from __future__ import absolute_import 21from __future__ import division 22from __future__ import print_function 23 24from tensorflow.python.ops import array_ops 25from tensorflow.python.ops import check_ops 26from tensorflow.python.ops import gen_ragged_math_ops 27from tensorflow.python.ops import math_ops 28 29 30 31def assert_splits_match(nested_splits_lists): 32 """Checks that the given splits lists are identical. 33 34 Performs static tests to ensure that the given splits lists are identical, 35 and returns a list of control dependency op tensors that check that they are 36 fully identical. 37 38 Args: 39 nested_splits_lists: A list of nested_splits_lists, where each split_list is 40 a list of `splits` tensors from a `RaggedTensor`, ordered from outermost 41 ragged dimension to innermost ragged dimension. 42 43 Returns: 44 A list of control dependency op tensors. 45 Raises: 46 ValueError: If the splits are not identical. 47 """ 48 error_msg = "Inputs must have identical ragged splits" 49 for splits_list in nested_splits_lists: 50 if len(splits_list) != len(nested_splits_lists[0]): 51 raise ValueError(error_msg) 52 return [ 53 check_ops.assert_equal(s1, s2, message=error_msg) 54 for splits_list in nested_splits_lists[1:] 55 for (s1, s2) in zip(nested_splits_lists[0], splits_list) 56 ] 57 58 59# Note: imported here to avoid circular dependency of array_ops. 60get_positive_axis = array_ops.get_positive_axis 61convert_to_int_tensor = array_ops.convert_to_int_tensor 62repeat = array_ops.repeat_with_axis 63 64 65def lengths_to_splits(lengths): 66 """Returns splits corresponding to the given lengths.""" 67 return array_ops.concat([[0], math_ops.cumsum(lengths)], axis=-1) 68 69 70def repeat_ranges(params, splits, repeats): 71 """Repeats each range of `params` (as specified by `splits`) `repeats` times. 72 73 Let the `i`th range of `params` be defined as 74 `params[splits[i]:splits[i + 1]]`. Then this function returns a tensor 75 containing range 0 repeated `repeats[0]` times, followed by range 1 repeated 76 `repeats[1]`, ..., followed by the last range repeated `repeats[-1]` times. 77 78 Args: 79 params: The `Tensor` whose values should be repeated. 80 splits: A splits tensor indicating the ranges of `params` that should be 81 repeated. 82 repeats: The number of times each range should be repeated. Supports 83 broadcasting from a scalar value. 84 85 Returns: 86 A `Tensor` with the same rank and type as `params`. 87 88 #### Example: 89 90 >>> print(repeat_ranges( 91 ... params=tf.constant(['a', 'b', 'c']), 92 ... splits=tf.constant([0, 2, 3]), 93 ... repeats=tf.constant(3))) 94 tf.Tensor([b'a' b'b' b'a' b'b' b'a' b'b' b'c' b'c' b'c'], 95 shape=(9,), dtype=string) 96 """ 97 # Divide `splits` into starts and limits, and repeat them `repeats` times. 98 if repeats.shape.ndims != 0: 99 repeated_starts = repeat(splits[:-1], repeats, axis=0) 100 repeated_limits = repeat(splits[1:], repeats, axis=0) 101 else: 102 # Optimization: we can just call repeat once, and then slice the result. 103 repeated_splits = repeat(splits, repeats, axis=0) 104 n_splits = array_ops.shape(repeated_splits, out_type=repeats.dtype)[0] 105 repeated_starts = repeated_splits[:n_splits - repeats] 106 repeated_limits = repeated_splits[repeats:] 107 108 # Get indices for each range from starts to limits, and use those to gather 109 # the values in the desired repetition pattern. 110 one = array_ops.ones((), repeated_starts.dtype) 111 offsets = gen_ragged_math_ops.ragged_range( 112 repeated_starts, repeated_limits, one) 113 return array_ops.gather(params, offsets.rt_dense_values) 114