Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/keras/src/utils/np_utils.py: 22%

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"""Numpy-related utilities.""" 

16 

17import numpy as np 

18 

19# isort: off 

20from tensorflow.python.util.tf_export import keras_export 

21 

22 

23@keras_export("keras.utils.to_categorical") 

24def to_categorical(y, num_classes=None, dtype="float32"): 

25 """Converts a class vector (integers) to binary class matrix. 

26 

27 E.g. for use with `categorical_crossentropy`. 

28 

29 Args: 

30 y: Array-like with class values to be converted into a matrix 

31 (integers from 0 to `num_classes - 1`). 

32 num_classes: Total number of classes. If `None`, this would be inferred 

33 as `max(y) + 1`. 

34 dtype: The data type expected by the input. Default: `'float32'`. 

35 

36 Returns: 

37 A binary matrix representation of the input as a NumPy array. The class 

38 axis is placed last. 

39 

40 Example: 

41 

42 >>> a = tf.keras.utils.to_categorical([0, 1, 2, 3], num_classes=4) 

43 >>> print(a) 

44 [[1. 0. 0. 0.] 

45 [0. 1. 0. 0.] 

46 [0. 0. 1. 0.] 

47 [0. 0. 0. 1.]] 

48 

49 >>> b = tf.constant([.9, .04, .03, .03, 

50 ... .3, .45, .15, .13, 

51 ... .04, .01, .94, .05, 

52 ... .12, .21, .5, .17], 

53 ... shape=[4, 4]) 

54 >>> loss = tf.keras.backend.categorical_crossentropy(a, b) 

55 >>> print(np.around(loss, 5)) 

56 [0.10536 0.82807 0.1011 1.77196] 

57 

58 >>> loss = tf.keras.backend.categorical_crossentropy(a, a) 

59 >>> print(np.around(loss, 5)) 

60 [0. 0. 0. 0.] 

61 """ 

62 y = np.array(y, dtype="int") 

63 input_shape = y.shape 

64 

65 # Shrink the last dimension if the shape is (..., 1). 

66 if input_shape and input_shape[-1] == 1 and len(input_shape) > 1: 

67 input_shape = tuple(input_shape[:-1]) 

68 

69 y = y.reshape(-1) 

70 if not num_classes: 

71 num_classes = np.max(y) + 1 

72 n = y.shape[0] 

73 categorical = np.zeros((n, num_classes), dtype=dtype) 

74 categorical[np.arange(n), y] = 1 

75 output_shape = input_shape + (num_classes,) 

76 categorical = np.reshape(categorical, output_shape) 

77 return categorical 

78 

79 

80@keras_export("keras.utils.to_ordinal") 

81def to_ordinal(y, num_classes=None, dtype="float32"): 

82 """Converts a class vector (integers) to an ordinal regression matrix. 

83 

84 This utility encodes class vector to ordinal regression/classification 

85 matrix where each sample is indicated by a row and rank of that sample is 

86 indicated by number of ones in that row. 

87 

88 Args: 

89 y: Array-like with class values to be converted into a matrix 

90 (integers from 0 to `num_classes - 1`). 

91 num_classes: Total number of classes. If `None`, this would be inferred 

92 as `max(y) + 1`. 

93 dtype: The data type expected by the input. Default: `'float32'`. 

94 

95 Returns: 

96 An ordinal regression matrix representation of the input as a NumPy 

97 array. The class axis is placed last. 

98 

99 Example: 

100 

101 >>> a = tf.keras.utils.to_ordinal([0, 1, 2, 3], num_classes=4) 

102 >>> print(a) 

103 [[0. 0. 0.] 

104 [1. 0. 0.] 

105 [1. 1. 0.] 

106 [1. 1. 1.]] 

107 """ 

108 y = np.array(y, dtype="int") 

109 input_shape = y.shape 

110 

111 # Shrink the last dimension if the shape is (..., 1). 

112 if input_shape and input_shape[-1] == 1 and len(input_shape) > 1: 

113 input_shape = tuple(input_shape[:-1]) 

114 

115 y = y.reshape(-1) 

116 if not num_classes: 

117 num_classes = np.max(y) + 1 

118 n = y.shape[0] 

119 range_values = np.arange(num_classes - 1) 

120 range_values = np.tile(np.expand_dims(range_values, 0), [n, 1]) 

121 ordinal = np.zeros((n, num_classes - 1), dtype=dtype) 

122 ordinal[range_values < np.expand_dims(y, -1)] = 1 

123 output_shape = input_shape + (num_classes - 1,) 

124 ordinal = np.reshape(ordinal, output_shape) 

125 return ordinal 

126 

127 

128@keras_export("keras.utils.normalize") 

129def normalize(x, axis=-1, order=2): 

130 """Normalizes a Numpy array. 

131 

132 Args: 

133 x: Numpy array to normalize. 

134 axis: axis along which to normalize. 

135 order: Normalization order (e.g. `order=2` for L2 norm). 

136 

137 Returns: 

138 A normalized copy of the array. 

139 """ 

140 l2 = np.atleast_1d(np.linalg.norm(x, order, axis)) 

141 l2[l2 == 0] = 1 

142 return x / np.expand_dims(l2, axis) 

143