Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/keras/src/feature_column/sequence_feature_column.py: 36%

45 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"""This API defines FeatureColumn for sequential input. 

16 

17NOTE: This API is a work in progress and will likely be changing frequently. 

18""" 

19 

20from __future__ import absolute_import 

21from __future__ import division 

22from __future__ import print_function 

23 

24import tensorflow.compat.v2 as tf 

25 

26from keras.src import backend 

27from keras.src.feature_column import base_feature_layer as kfc 

28 

29# isort: off 

30from tensorflow.python.util.tf_export import keras_export 

31 

32 

33@keras_export("keras.experimental.SequenceFeatures") 

34class SequenceFeatures(kfc._BaseFeaturesLayer): 

35 """A layer for sequence input. 

36 

37 All `feature_columns` must be sequence dense columns with the same 

38 `sequence_length`. The output of this method can be fed into sequence 

39 networks, such as RNN. 

40 

41 The output of this method is a 3D `Tensor` of shape `[batch_size, T, D]`. 

42 `T` is the maximum sequence length for this batch, which could differ from 

43 batch to batch. 

44 

45 If multiple `feature_columns` are given with `Di` `num_elements` each, their 

46 outputs are concatenated. So, the final `Tensor` has shape 

47 `[batch_size, T, D0 + D1 + ... + Dn]`. 

48 

49 Example: 

50 

51 ```python 

52 

53 import tensorflow as tf 

54 

55 # Behavior of some cells or feature columns may depend on whether we are in 

56 # training or inference mode, e.g. applying dropout. 

57 training = True 

58 rating = tf.feature_column.sequence_numeric_column('rating') 

59 watches = tf.feature_column.sequence_categorical_column_with_identity( 

60 'watches', num_buckets=1000) 

61 watches_embedding = tf.feature_column.embedding_column(watches, 

62 dimension=10) 

63 columns = [rating, watches_embedding] 

64 

65 features = { 

66 'rating': tf.sparse.from_dense([[1.0,1.1, 0, 0, 0], 

67 [2.0,2.1,2.2, 2.3, 2.5]]), 

68 'watches': tf.sparse.from_dense([[2, 85, 0, 0, 0],[33,78, 2, 73, 1]]) 

69 } 

70 

71 sequence_input_layer = tf.keras.experimental.SequenceFeatures(columns) 

72 sequence_input, sequence_length = sequence_input_layer( 

73 features, training=training) 

74 sequence_length_mask = tf.sequence_mask(sequence_length) 

75 hidden_size = 32 

76 rnn_cell = tf.keras.layers.SimpleRNNCell(hidden_size) 

77 rnn_layer = tf.keras.layers.RNN(rnn_cell) 

78 outputs, state = rnn_layer(sequence_input, mask=sequence_length_mask) 

79 ``` 

80 """ 

81 

82 def __init__(self, feature_columns, trainable=True, name=None, **kwargs): 

83 """ "Constructs a SequenceFeatures layer. 

84 

85 Args: 

86 feature_columns: An iterable of dense sequence columns. Valid columns 

87 are 

88 - `embedding_column` that wraps a 

89 `sequence_categorical_column_with_*` 

90 - `sequence_numeric_column`. 

91 trainable: Boolean, whether the layer's variables will be updated via 

92 gradient descent during training. 

93 name: Name to give to the SequenceFeatures. 

94 **kwargs: Keyword arguments to construct a layer. 

95 

96 Raises: 

97 ValueError: If any of the `feature_columns` is not a 

98 `SequenceDenseColumn`. 

99 """ 

100 super().__init__( 

101 feature_columns=feature_columns, 

102 trainable=trainable, 

103 name=name, 

104 expected_column_type=tf.__internal__.feature_column.SequenceDenseColumn, # noqa: E501 

105 **kwargs 

106 ) 

107 

108 @property 

109 def _is_feature_layer(self): 

110 return True 

111 

112 def _target_shape(self, input_shape, total_elements): 

113 return (input_shape[0], input_shape[1], total_elements) 

114 

115 def call(self, features, training=None): 

116 """Returns sequence input corresponding to the `feature_columns`. 

117 

118 Args: 

119 features: A dict mapping keys to tensors. 

120 training: Python boolean or None, indicating whether to the layer is 

121 being run in training mode. This argument is passed to the call 

122 method of any `FeatureColumn` that takes a `training` argument. For 

123 example, if a `FeatureColumn` performed dropout, the column could 

124 expose a `training` argument to control whether the dropout should 

125 be applied. If `None`, defaults to 

126 `tf.keras.backend.learning_phase()`. 

127 

128 

129 Returns: 

130 An `(input_layer, sequence_length)` tuple where: 

131 - input_layer: A float `Tensor` of shape `[batch_size, T, D]`. 

132 `T` is the maximum sequence length for this batch, which could 

133 differ from batch to batch. `D` is the sum of `num_elements` for 

134 all `feature_columns`. 

135 - sequence_length: An int `Tensor` of shape `[batch_size]`. The 

136 sequence length for each example. 

137 

138 Raises: 

139 ValueError: If features are not a dictionary. 

140 """ 

141 if not isinstance(features, dict): 

142 raise ValueError( 

143 "We expected a dictionary here. Instead we got: ", features 

144 ) 

145 if training is None: 

146 training = backend.learning_phase() 

147 transformation_cache = ( 

148 tf.__internal__.feature_column.FeatureTransformationCache(features) 

149 ) 

150 output_tensors = [] 

151 sequence_lengths = [] 

152 

153 for column in self._feature_columns: 

154 with backend.name_scope(column.name): 

155 try: 

156 ( 

157 dense_tensor, 

158 sequence_length, 

159 ) = column.get_sequence_dense_tensor( 

160 transformation_cache, 

161 self._state_manager, 

162 training=training, 

163 ) 

164 except TypeError: 

165 ( 

166 dense_tensor, 

167 sequence_length, 

168 ) = column.get_sequence_dense_tensor( 

169 transformation_cache, self._state_manager 

170 ) 

171 # Flattens the final dimension to produce a 3D Tensor. 

172 output_tensors.append( 

173 self._process_dense_tensor(column, dense_tensor) 

174 ) 

175 sequence_lengths.append(sequence_length) 

176 

177 # Check and process sequence lengths. 

178 kfc._verify_static_batch_size_equality( 

179 sequence_lengths, self._feature_columns 

180 ) 

181 sequence_length = _assert_all_equal_and_return(sequence_lengths) 

182 

183 return self._verify_and_concat_tensors(output_tensors), sequence_length 

184 

185 

186def _assert_all_equal_and_return(tensors, name=None): 

187 """Asserts that all tensors are equal and returns the first one.""" 

188 with backend.name_scope(name or "assert_all_equal"): 

189 if len(tensors) == 1: 

190 return tensors[0] 

191 assert_equal_ops = [] 

192 for t in tensors[1:]: 

193 assert_equal_ops.append(tf.compat.v1.assert_equal(tensors[0], t)) 

194 with tf.control_dependencies(assert_equal_ops): 

195 return tf.identity(tensors[0]) 

196