xref: /aosp_15_r20/external/tensorflow/tensorflow/python/ops/ragged/segment_id_ops.py (revision b6fb3261f9314811a0f4371741dbb8839866f948) 
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"""Ops for converting between row_splits and segment_ids."""
16
17from tensorflow.python.framework import dtypes
18from tensorflow.python.framework import ops
19from tensorflow.python.framework import tensor_shape
20from tensorflow.python.framework import tensor_util
21from tensorflow.python.ops import array_ops
22from tensorflow.python.ops import math_ops
23from tensorflow.python.ops.ragged import ragged_util
24from tensorflow.python.util import dispatch
25from tensorflow.python.util.tf_export import tf_export
26
27
28# For background on "segments" and "segment ids", see:
29# https://www.tensorflow.org/api_docs/python/tf/math#Segmentation
30@tf_export("ragged.row_splits_to_segment_ids")
31@dispatch.add_dispatch_support
32def row_splits_to_segment_ids(splits, name=None, out_type=None):
33  """Generates the segmentation corresponding to a RaggedTensor `row_splits`.
34
35  Returns an integer vector `segment_ids`, where `segment_ids[i] == j` if
36  `splits[j] <= i < splits[j+1]`.  Example:
37
38  >>> print(tf.ragged.row_splits_to_segment_ids([0, 3, 3, 5, 6, 9]))
39   tf.Tensor([0 0 0 2 2 3 4 4 4], shape=(9,), dtype=int64)
40
41  Args:
42    splits: A sorted 1-D integer Tensor.  `splits[0]` must be zero.
43    name: A name prefix for the returned tensor (optional).
44    out_type: The dtype for the return value.  Defaults to `splits.dtype`,
45      or `tf.int64` if `splits` does not have a dtype.
46
47  Returns:
48    A sorted 1-D integer Tensor, with `shape=[splits[-1]]`
49
50  Raises:
51    ValueError: If `splits` is invalid.
52  """
53  with ops.name_scope(name, "RaggedSplitsToSegmentIds", [splits]) as name:
54    splits = ops.convert_to_tensor(
55        splits, name="splits",
56        preferred_dtype=dtypes.int64)
57    if splits.dtype not in (dtypes.int32, dtypes.int64):
58      raise ValueError("splits must have dtype int32 or int64")
59    splits.shape.assert_has_rank(1)
60    if tensor_shape.dimension_value(splits.shape[0]) == 0:
61      raise ValueError("Invalid row_splits: []")
62    if out_type is None:
63      out_type = splits.dtype
64    else:
65      out_type = dtypes.as_dtype(out_type)
66    row_lengths = splits[1:] - splits[:-1]
67    nrows = array_ops.shape(splits, out_type=out_type)[-1] - 1
68    indices = math_ops.range(nrows)
69    return ragged_util.repeat(indices, repeats=row_lengths, axis=0)
70
71
72# For background on "segments" and "segment ids", see:
73# https://www.tensorflow.org/api_docs/python/tf/math#Segmentation
74@tf_export("ragged.segment_ids_to_row_splits")
75@dispatch.add_dispatch_support
76def segment_ids_to_row_splits(segment_ids, num_segments=None,
77                              out_type=None, name=None):
78  """Generates the RaggedTensor `row_splits` corresponding to a segmentation.
79
80  Returns an integer vector `splits`, where `splits[0] = 0` and
81  `splits[i] = splits[i-1] + count(segment_ids==i)`.  Example:
82
83  >>> print(tf.ragged.segment_ids_to_row_splits([0, 0, 0, 2, 2, 3, 4, 4, 4]))
84  tf.Tensor([0 3 3 5 6 9], shape=(6,), dtype=int64)
85
86  Args:
87    segment_ids: A 1-D integer Tensor.
88    num_segments: A scalar integer indicating the number of segments.  Defaults
89      to `max(segment_ids) + 1` (or zero if `segment_ids` is empty).
90    out_type: The dtype for the return value.  Defaults to `segment_ids.dtype`,
91      or `tf.int64` if `segment_ids` does not have a dtype.
92    name: A name prefix for the returned tensor (optional).
93
94  Returns:
95    A sorted 1-D integer Tensor, with `shape=[num_segments + 1]`.
96  """
97  # Local import bincount_ops to avoid import-cycle.
98  from tensorflow.python.ops import bincount_ops  # pylint: disable=g-import-not-at-top
99  if out_type is None:
100    if isinstance(segment_ids, ops.Tensor):
101      out_type = segment_ids.dtype
102    elif isinstance(num_segments, ops.Tensor):
103      out_type = num_segments.dtype
104    else:
105      out_type = dtypes.int64
106  else:
107    out_type = dtypes.as_dtype(out_type)
108  with ops.name_scope(name, "SegmentIdsToRaggedSplits", [segment_ids]) as name:
109    # Note: we cast int64 tensors to int32, since bincount currently only
110    # supports int32 inputs.
111    segment_ids = ragged_util.convert_to_int_tensor(segment_ids, "segment_ids",
112                                                    dtype=dtypes.int32)
113    segment_ids.shape.assert_has_rank(1)
114    if num_segments is not None:
115      num_segments = ragged_util.convert_to_int_tensor(num_segments,
116                                                       "num_segments",
117                                                       dtype=dtypes.int32)
118      num_segments.shape.assert_has_rank(0)
119
120    row_lengths = bincount_ops.bincount(
121        segment_ids,
122        minlength=num_segments,
123        maxlength=num_segments,
124        dtype=out_type)
125    splits = array_ops.concat([[0], math_ops.cumsum(row_lengths)], axis=0)
126
127    # Update shape information, if possible.
128    if num_segments is not None:
129      const_num_segments = tensor_util.constant_value(num_segments)
130      if const_num_segments is not None:
131        splits.set_shape(tensor_shape.TensorShape([const_num_segments + 1]))
132
133    return splits
134