Coverage for /pythoncovmergedfiles/medio/medio/usr/local/lib/python3.8/site-packages/tensorflow_addons/activations/rrelu.py: 31%

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 

16import tensorflow as tf 

17from tensorflow_addons.utils.types import Number, TensorLike 

18from typing import Optional 

19 

20 

21@tf.keras.utils.register_keras_serializable(package="Addons") 

22def rrelu( 

23 x: TensorLike, 

24 lower: Number = 0.125, 

25 upper: Number = 0.3333333333333333, 

26 training: Optional[bool] = None, 

27 seed: Optional[int] = None, 

28 rng: Optional[tf.random.Generator] = None, 

29) -> tf.Tensor: 

30 r"""Randomized leaky rectified liner unit function. 

31 

32 Computes rrelu function: 

33 

34 $$ 

35 \mathrm{rrelu}(x) = 

36 \begin{cases} 

37 x & \text{if } x > 0 \\ 

38 a x 

39 \end{cases}, 

40 $$ 

41 

42 where 

43 

44 $$ 

45 a \sim \mathcal{U}(\mathrm{lower}, \mathrm{upper}) 

46 $$ 

47 

48 when `training` is `True`; or 

49 

50 $$ 

51 a = \frac{\mathrm{lower} + \mathrm{upper}}{2} 

52 $$ 

53 

54 when `training` is `False`. 

55 

56 See [Empirical Evaluation of Rectified Activations in Convolutional Network](https://arxiv.org/abs/1505.00853). 

57 

58 Usage: 

59 

60 >>> x = tf.constant([-1.0, 0.0, 1.0]) 

61 >>> tfa.activations.rrelu(x, training=False) 

62 <tf.Tensor: shape=(3,), dtype=float32, numpy=array([-0.22916667, 0. , 1. ], dtype=float32)> 

63 >>> tfa.activations.rrelu(x, training=True, seed=2020) 

64 <tf.Tensor: shape=(3,), dtype=float32, numpy=array([-0.22631127, 0. , 1. ], dtype=float32)> 

65 >>> generator = tf.random.Generator.from_seed(2021) 

66 >>> tfa.activations.rrelu(x, training=True, rng=generator) 

67 <tf.Tensor: shape=(3,), dtype=float32, numpy=array([-0.16031083, 0. , 1. ], dtype=float32)> 

68 

69 Args: 

70 x: A `Tensor`. Must be one of the following types: 

71 `bfloat16`, `float16`, `float32`, `float64`. 

72 lower: `float`, lower bound for random alpha. 

73 upper: `float`, upper bound for random alpha. 

74 training: `bool`, indicating whether the `call` 

75 is meant for training or inference. 

76 seed: `int`, this sets the operation-level seed. 

77 rng: A `tf.random.Generator`. 

78 Returns: 

79 result: A `Tensor`. Has the same type as `x`. 

80 """ 

81 x = tf.convert_to_tensor(x) 

82 lower = tf.cast(lower, x.dtype) 

83 upper = tf.cast(upper, x.dtype) 

84 

85 def random_a(): 

86 if rng is not None and seed is not None: 

87 raise ValueError( 

88 "Either seed or rng should be specified. Not both at the same time." 

89 ) 

90 

91 if rng is not None: 

92 return rng.uniform(tf.shape(x), minval=lower, maxval=upper, dtype=x.dtype) 

93 

94 return tf.random.uniform( 

95 tf.shape(x), minval=lower, maxval=upper, dtype=x.dtype, seed=seed 

96 ) 

97 

98 a = tf.keras.backend.in_train_phase(random_a, (lower + upper) / 2, training) 

99 

100 return tf.where(x >= 0, x, a * x)