Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/tensorflow/python/ops/distributions/exponential.py: 61%

1# Copyright 2016 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"""The Exponential distribution class.""" 

16 

17import numpy as np 

18 

19from tensorflow.python.framework import dtypes 

20from tensorflow.python.framework import ops 

21from tensorflow.python.ops import array_ops 

22from tensorflow.python.ops import math_ops 

23from tensorflow.python.ops import nn 

24from tensorflow.python.ops import random_ops 

25from tensorflow.python.ops.distributions import gamma 

26from tensorflow.python.util import deprecation 

27from tensorflow.python.util.tf_export import tf_export 

28 

29 

30__all__ = [ 

31 "Exponential", 

32 "ExponentialWithSoftplusRate", 

33] 

34 

35 

36@tf_export(v1=["distributions.Exponential"]) 

37class Exponential(gamma.Gamma): 

38 """Exponential distribution. 

39 

40 The Exponential distribution is parameterized by an event `rate` parameter. 

41 

42 #### Mathematical Details 

43 

44 The probability density function (pdf) is, 

45 

46 ```none 

47 pdf(x; lambda, x > 0) = exp(-lambda x) / Z 

48 Z = 1 / lambda 

49 ``` 

50 

51 where `rate = lambda` and `Z` is the normalizaing constant. 

52 

53 The Exponential distribution is a special case of the Gamma distribution, 

54 i.e., 

55 

56 ```python 

57 Exponential(rate) = Gamma(concentration=1., rate) 

58 ``` 

59 

60 The Exponential distribution uses a `rate` parameter, or "inverse scale", 

61 which can be intuited as, 

62 

63 ```none 

64 X ~ Exponential(rate=1) 

65 Y = X / rate 

66 ``` 

67 

68 """ 

69 

70 @deprecation.deprecated( 

71 "2019-01-01", 

72 "The TensorFlow Distributions library has moved to " 

73 "TensorFlow Probability " 

74 "(https://github.com/tensorflow/probability). You " 

75 "should update all references to use `tfp.distributions` " 

76 "instead of `tf.distributions`.", 

77 warn_once=True) 

78 def __init__(self, 

79 rate, 

80 validate_args=False, 

81 allow_nan_stats=True, 

82 name="Exponential"): 

83 """Construct Exponential distribution with parameter `rate`. 

84 

85 Args: 

86 rate: Floating point tensor, equivalent to `1 / mean`. Must contain only 

87 positive values. 

88 validate_args: Python `bool`, default `False`. When `True` distribution 

89 parameters are checked for validity despite possibly degrading runtime 

90 performance. When `False` invalid inputs may silently render incorrect 

91 outputs. 

92 allow_nan_stats: Python `bool`, default `True`. When `True`, statistics 

93 (e.g., mean, mode, variance) use the value "`NaN`" to indicate the 

94 result is undefined. When `False`, an exception is raised if one or 

95 more of the statistic's batch members are undefined. 

96 name: Python `str` name prefixed to Ops created by this class. 

97 """ 

98 parameters = dict(locals()) 

99 # Even though all statistics of are defined for valid inputs, this is not 

100 # true in the parent class "Gamma." Therefore, passing 

101 # allow_nan_stats=True 

102 # through to the parent class results in unnecessary asserts. 

103 with ops.name_scope(name, values=[rate]) as name: 

104 self._rate = ops.convert_to_tensor(rate, name="rate") 

105 super(Exponential, self).__init__( 

106 concentration=array_ops.ones([], dtype=self._rate.dtype), 

107 rate=self._rate, 

108 allow_nan_stats=allow_nan_stats, 

109 validate_args=validate_args, 

110 name=name) 

111 self._parameters = parameters 

112 self._graph_parents += [self._rate] 

113 

114 @staticmethod 

115 def _param_shapes(sample_shape): 

116 return {"rate": ops.convert_to_tensor(sample_shape, dtype=dtypes.int32)} 

117 

118 @property 

119 def rate(self): 

120 return self._rate 

121 

122 def _log_survival_function(self, value): 

123 return self._log_prob(value) - math_ops.log(self._rate) 

124 

125 def _sample_n(self, n, seed=None): 

126 shape = array_ops.concat([[n], array_ops.shape(self._rate)], 0) 

127 # Uniform variates must be sampled from the open-interval `(0, 1)` rather 

128 # than `[0, 1)`. To do so, we use `np.finfo(self.dtype.as_numpy_dtype).tiny` 

129 # because it is the smallest, positive, "normal" number. A "normal" number 

130 # is such that the mantissa has an implicit leading 1. Normal, positive 

131 # numbers x, y have the reasonable property that, `x + y >= max(x, y)`. In 

132 # this case, a subnormal number (i.e., np.nextafter) can cause us to sample 

133 # 0. 

134 sampled = random_ops.random_uniform( 

135 shape, 

136 minval=np.finfo(self.dtype.as_numpy_dtype).tiny, 

137 maxval=1., 

138 seed=seed, 

139 dtype=self.dtype) 

140 return -math_ops.log(sampled) / self._rate 

141 

142 

143class ExponentialWithSoftplusRate(Exponential): 

144 """Exponential with softplus transform on `rate`.""" 

145 

146 @deprecation.deprecated( 

147 "2019-01-01", 

148 "Use `tfd.Exponential(tf.nn.softplus(rate)).", 

149 warn_once=True) 

150 def __init__(self, 

151 rate, 

152 validate_args=False, 

153 allow_nan_stats=True, 

154 name="ExponentialWithSoftplusRate"): 

155 parameters = dict(locals()) 

156 with ops.name_scope(name, values=[rate]) as name: 

157 super(ExponentialWithSoftplusRate, self).__init__( 

158 rate=nn.softplus(rate, name="softplus_rate"), 

159 validate_args=validate_args, 

160 allow_nan_stats=allow_nan_stats, 

161 name=name) 

162 self._parameters = parameters