xref: /aosp_15_r20/external/tensorflow/tensorflow/python/keras/utils/version_utils.py (revision b6fb3261f9314811a0f4371741dbb8839866f948) 
1# Copyright 2019 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# pylint: disable=protected-access
16"""Utilities for Keras classes with v1 and v2 versions."""
17
18from tensorflow.python.eager import context
19from tensorflow.python.framework import ops
20from tensorflow.python.keras.utils.generic_utils import LazyLoader
21
22# TODO(b/134426265): Switch back to single-quotes once the issue
23# with copybara is fixed.
24# pylint: disable=g-inconsistent-quotes
25training = LazyLoader(
26    "training", globals(),
27    "tensorflow.python.keras.engine.training")
28training_v1 = LazyLoader(
29    "training_v1", globals(),
30    "tensorflow.python.keras.engine.training_v1")
31base_layer = LazyLoader(
32    "base_layer", globals(),
33    "tensorflow.python.keras.engine.base_layer")
34base_layer_v1 = LazyLoader(
35    "base_layer_v1", globals(),
36    "tensorflow.python.keras.engine.base_layer_v1")
37callbacks = LazyLoader(
38    "callbacks", globals(),
39    "tensorflow.python.keras.callbacks")
40callbacks_v1 = LazyLoader(
41    "callbacks_v1", globals(),
42    "tensorflow.python.keras.callbacks_v1")
43
44
45# pylint: enable=g-inconsistent-quotes
46
47
48class ModelVersionSelector(object):
49  """Chooses between Keras v1 and v2 Model class."""
50
51  def __new__(cls, *args, **kwargs):  # pylint: disable=unused-argument
52    use_v2 = should_use_v2()
53    cls = swap_class(cls, training.Model, training_v1.Model, use_v2)  # pylint: disable=self-cls-assignment
54    return super(ModelVersionSelector, cls).__new__(cls)
55
56
57class LayerVersionSelector(object):
58  """Chooses between Keras v1 and v2 Layer class."""
59
60  def __new__(cls, *args, **kwargs):  # pylint: disable=unused-argument
61    use_v2 = should_use_v2()
62    cls = swap_class(cls, base_layer.Layer, base_layer_v1.Layer, use_v2)  # pylint: disable=self-cls-assignment
63    return super(LayerVersionSelector, cls).__new__(cls)
64
65
66class TensorBoardVersionSelector(object):
67  """Chooses between Keras v1 and v2 TensorBoard callback class."""
68
69  def __new__(cls, *args, **kwargs):  # pylint: disable=unused-argument
70    use_v2 = should_use_v2()
71    start_cls = cls
72    cls = swap_class(start_cls, callbacks.TensorBoard, callbacks_v1.TensorBoard,
73                     use_v2)
74    if start_cls == callbacks_v1.TensorBoard and cls == callbacks.TensorBoard:
75      # Since the v2 class is not a subclass of the v1 class, __init__ has to
76      # be called manually.
77      return cls(*args, **kwargs)
78    return super(TensorBoardVersionSelector, cls).__new__(cls)
79
80
81def should_use_v2():
82  """Determine if v1 or v2 version should be used."""
83  if context.executing_eagerly():
84    return True
85  elif ops.executing_eagerly_outside_functions():
86    # Check for a v1 `wrap_function` FuncGraph.
87    # Code inside a `wrap_function` is treated like v1 code.
88    graph = ops.get_default_graph()
89    if (getattr(graph, "name", False) and
90        graph.name.startswith("wrapped_function")):
91      return False
92    return True
93  else:
94    return False
95
96
97def swap_class(cls, v2_cls, v1_cls, use_v2):
98  """Swaps in v2_cls or v1_cls depending on graph mode."""
99  if cls == object:
100    return cls
101  if cls in (v2_cls, v1_cls):
102    return v2_cls if use_v2 else v1_cls
103
104  # Recursively search superclasses to swap in the right Keras class.
105  new_bases = []
106  for base in cls.__bases__:
107    if ((use_v2 and issubclass(base, v1_cls)
108         # `v1_cls` often extends `v2_cls`, so it may still call `swap_class`
109         # even if it doesn't need to. That being said, it may be the safest
110         # not to over optimize this logic for the sake of correctness,
111         # especially if we swap v1 & v2 classes that don't extend each other,
112         # or when the inheritance order is different.
113         or (not use_v2 and issubclass(base, v2_cls)))):
114      new_base = swap_class(base, v2_cls, v1_cls, use_v2)
115    else:
116      new_base = base
117    new_bases.append(new_base)
118  cls.__bases__ = tuple(new_bases)
119  return cls
120
121
122def disallow_legacy_graph(cls_name, method_name):
123  if not ops.executing_eagerly_outside_functions():
124    error_msg = (
125        "Calling `{cls_name}.{method_name}` in graph mode is not supported "
126        "when the `{cls_name}` instance was constructed with eager mode "
127        "enabled. Please construct your `{cls_name}` instance in graph mode or"
128        " call `{cls_name}.{method_name}` with eager mode enabled.")
129    error_msg = error_msg.format(cls_name=cls_name, method_name=method_name)
130    raise ValueError(error_msg)
131
132
133def is_v1_layer_or_model(obj):
134  return isinstance(obj, (base_layer_v1.Layer, training_v1.Model))
135