/
cells.py
185 lines (165 loc) · 8.71 KB
/
cells.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
from keras.models import Model
from keras import initializers
from keras import constraints
from keras import regularizers
from keras.layers import *
from .engine import RNNCell
def _slice(x, dim, index):
return x[:, index * dim: dim * (index + 1)]
def get_slices(x, n):
dim = int(K.int_shape(x)[1] / n)
return [Lambda(_slice, arguments={'dim': dim, 'index': i}, output_shape=lambda s: (s[0], dim))(x) for i in range(n)]
class Identity(Layer):
def call(self, x):
return x + 0.
class ExtendedRNNCell(RNNCell):
def __init__(self, units=None,
activation='tanh',
recurrent_activation='hard_sigmoid',
use_bias=True,
kernel_initializer='glorot_uniform',
recurrent_initializer='orthogonal',
bias_initializer='zeros',
kernel_regularizer=None,
recurrent_regularizer=None,
bias_regularizer=None,
activity_regularizer=None,
kernel_constraint=None,
recurrent_constraint=None,
bias_constraint=None,
**kwargs):
if units is None:
assert 'output_dim' in kwargs, 'Missing argument: units'
else:
kwargs['output_dim'] = units
self.activation = activations.get(activation)
self.recurrent_activation = activations.get(recurrent_activation)
self.use_bias = use_bias
self.kernel_initializer = initializers.get(kernel_initializer)
self.recurrent_initializer = initializers.get(recurrent_initializer)
self.bias_initializer = initializers.get(bias_initializer)
self.kernel_regularizer = regularizers.get(kernel_regularizer)
self.recurrent_regularizer = regularizers.get(recurrent_regularizer)
self.bias_regularizer = regularizers.get(bias_regularizer)
self.activity_regularizer = regularizers.get(activity_regularizer)
self.kernel_constraint = constraints.get(kernel_constraint)
self.recurrent_constraint = constraints.get(recurrent_constraint)
self.bias_constraint = constraints.get(bias_constraint)
super(ExtendedRNNCell, self).__init__(**kwargs)
def get_config(self):
config = {
'activation': activations.serialize(self.activation),
'recurrent_activation': activations.serialize(self.recurrent_activation),
'use_bias': self.use_bias,
'kernel_initializer': initializers.serialize(self.kernel_initializer),
'recurrent_initializer': initializers.serialize(self.recurrent_initializer),
'bias_initializer': initializers.serialize(self.bias_initializer),
'kernel_regularizer': regularizers.serialize(self.kernel_regularizer),
'recurrent_regularizer': regularizers.serialize(self.recurrent_regularizer),
'bias_regularizer': regularizers.serialize(self.bias_regularizer),
'activity_regularizer': regularizers.serialize(self.activity_regularizer),
'kernel_constraint': constraints.serialize(self.kernel_constraint),
'recurrent_constraint': constraints.serialize(self.recurrent_constraint),
'bias_constraint': constraints.serialize(self.bias_constraint)
}
base_config = super(ExtendedRNNCell, self).get_config()
config.update(base_config)
return config
class SimpleRNNCell(ExtendedRNNCell):
def build_model(self, input_shape):
output_dim = self.output_dim
output_shape = (input_shape[0], output_dim)
x = Input(batch_shape=input_shape)
h_tm1 = Input(batch_shape=output_shape)
kernel = Dense(output_dim,
kernel_initializer=self.kernel_initializer,
kernel_regularizer=self.kernel_regularizer,
kernel_constraint=self.kernel_constraint,
use_bias=self.use_bias,
bias_initializer=self.bias_initializer,
bias_regularizer=self.bias_regularizer,
bias_constraint=self.bias_constraint)
recurrent_kernel = Dense(output_dim,
kernel_initializer=self.recurrent_initializer,
kernel_regularizer=self.recurrent_regularizer,
kernel_constraint=self.recurrent_constraint,
use_bias=False)
h = add([kernel(x), recurrent_kernel(h_tm1)])
h = Activation(self.activation)(h)
return Model([x, h_tm1], [h, Identity()(h)])
class GRUCell(ExtendedRNNCell):
def build_model(self, input_shape):
output_dim = self.output_dim
input_dim = input_shape[-1]
output_shape = (input_shape[0], output_dim)
x = Input(batch_shape=input_shape)
h_tm1 = Input(batch_shape=output_shape)
kernel = Dense(output_dim * 3,
kernel_initializer=self.kernel_initializer,
kernel_regularizer=self.kernel_regularizer,
kernel_constraint=self.kernel_constraint,
use_bias=self.use_bias,
bias_initializer=self.bias_initializer,
bias_regularizer=self.bias_regularizer,
bias_constraint=self.bias_constraint)
recurrent_kernel_1 = Dense(output_dim * 2,
kernel_initializer=self.recurrent_initializer,
kernel_regularizer=self.recurrent_regularizer,
kernel_constraint=self.recurrent_constraint,
use_bias=False)
recurrent_kernel_2 = Dense(output_dim,
kernel_initializer=self.recurrent_initializer,
kernel_regularizer=self.recurrent_regularizer,
kernel_constraint=self.recurrent_constraint,
use_bias=False)
kernel_out = kernel(x)
recurrent_kernel_1_out = recurrent_kernel_1(h_tm1)
x_z, x_r, x_h = get_slices(kernel_out, 3)
r_z, r_r = get_slices(recurrent_kernel_1_out, 2)
z = add([x_z, r_z])
z = Activation(self.recurrent_activation)(z) # sigma_g
r = add([x_r, r_r])
r = Activation(self.recurrent_activation)(r) # sigma_g
h_prime = add([recurrent_kernel_2(multiply([r, h_tm1])), x_h])
h_prime = Activation(self.activation)(h_prime) # sigma_h
# h = z * h' + (1 - z) * h_tm1
gate = Lambda(lambda x: x[0] * x[1] + (1. - x[0]) * x[2], output_shape=lambda s: s[0])
h = gate([z, h_prime, h_tm1])
return Model([x, h_tm1], [h, Identity()(h)])
class LSTMCell(ExtendedRNNCell):
def build_model(self, input_shape):
output_dim = self.output_dim
input_dim = input_shape[-1]
output_shape = (input_shape[0], output_dim)
x = Input(batch_shape=input_shape)
h_tm1 = Input(batch_shape=output_shape)
c_tm1 = Input(batch_shape=output_shape)
kernel = Dense(output_dim * 4,
kernel_initializer=self.kernel_initializer,
kernel_regularizer=self.kernel_regularizer,
kernel_constraint=self.kernel_constraint,
use_bias=self.use_bias,
bias_initializer=self.bias_initializer,
bias_regularizer=self.bias_regularizer,
bias_constraint=self.bias_constraint)
recurrent_kernel = Dense(output_dim * 4,
kernel_initializer=self.recurrent_initializer,
kernel_regularizer=self.recurrent_regularizer,
kernel_constraint=self.recurrent_constraint,
use_bias=False)
kernel_out = kernel(x)
recurrent_kernel_out = recurrent_kernel(h_tm1)
x_f, x_i, x_o, x_c = get_slices(kernel_out, 4)
r_f, r_i, r_o, r_c = get_slices(recurrent_kernel_out, 4)
f = add([x_f, r_f])
f = Activation(self.recurrent_activation)(f)
i = add([x_i, r_i])
i = Activation(self.recurrent_activation)(i)
o = add([x_o, r_o])
o = Activation(self.recurrent_activation)(o)
c_prime = add([x_c, r_c])
c_prime = Activation(self.activation)(c_prime)
c = add([multiply([f, c_tm1]), multiply([i, c_prime])])
c = Activation(self.activation)(c)
h = multiply([o, c])
return Model([x, h_tm1, c_tm1], [h, Identity()(h), c])