-
Notifications
You must be signed in to change notification settings - Fork 1
/
layers.py
144 lines (114 loc) · 4.52 KB
/
layers.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
import os
import tensorflow as tf
from tensorflow.keras.layers import Layer
from tensorflow.keras.models import Model
from tensorflow.keras.regularizers import l2
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
class GConv(Layer):
def __init__(self, units, dropout=False, activation='relu',
kernel_init='he_normal', bias_init='random_normal', **kwargs):
super(GConv, self).__init__()
self.units = units
self.dropout = dropout
self.kernel_init = kernel_init
self.bias_init = bias_init
self.act_name = activation
self.act = tf.keras.layers.Activation(activation)
def build(self, input_shape):
# print(input_shape)
input_dim = input_shape[0][-1]
self.w = self.add_weight(
name='w',
shape=(input_dim, self.units),
initializer=self.kernel_init,
regularizer=l2(0.001),
trainable=True
)
self.b = self.add_weight(
name='b',
shape=(self.units,), initializer=self.bias_init, trainable=True
)
def call(self, inputs):
x, a = inputs
if self.dropout:
x = tf.nn.dropout(x, rate=0.2, seed=1)
output = tf.matmul(x, self.w)
output = tf.sparse.sparse_dense_matmul(a, output)
# output = K.bias_add(output, self.b)
output = self.act(output + self.b)
return output
def get_config(self):
config = {
'units':
self.units,
}
base_config = super(GConv, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
class ConcatAdj(Layer):
def __init__(self, block_diag=True, **kwargs):
super(ConcatAdj, self).__init__()
self.block_diag = block_diag
def call(self, a1, a2):
M, N = a1.shape[0], a2.shape[0]
new_inds = tf.concat((a1.indices,
tf.add(a2.indices, tf.constant(M, dtype=tf.int64))), 0)
new_vals = tf.concat((a1.values, a2.values), -1)
a_out = tf.sparse.SparseTensor(indices=new_inds,
values=new_vals, dense_shape=(M + N, M + N))
return a_out
def get_config(self):
config = {
'block_diag':
self.block_diag,
}
base_config = super(ConcatAdj, self).get_config()
return dict(list(base_config.items()) + list(config.items()))
class MLPBlock(Model):
def __init__(self, filters):
super(MLPBlock, self).__init__()
filters1, filters2, filters3 = filters
self.conv2a = tf.keras.layers.Conv1D(filters1, 1)
# self.bn2a = tf.keras.layers.BatchNormalization()
self.conv2b = tf.keras.layers.Conv1D(filters2, 1)
# self.bn2b = tf.keras.layers.BatchNormalization()
self.conv2c = tf.keras.layers.Conv1D(filters3, 1)
# self.bn2c = tf.keras.layers.BatchNormalization()
def call(self, input_tensor, training=False):
x = tf.expand_dims(input_tensor, 0)
x = self.conv2a(x)
# x = self.bn2a(x, training=training)
x = tf.nn.relu(x)
x = self.conv2b(x)
# x = self.bn2b(x, training=training)
x = tf.nn.relu(x)
x = self.conv2c(x)
# x = self.bn2c(x, training=training)
return tf.nn.softmax(x)
class CyclicalLR(tf.keras.optimizers.schedules.LearningRateSchedule):
def __init__(self, base_lr=0.001, max_lr=0.01, step_size=2000., name=None):
self.base_lr = base_lr
self.max_lr = max_lr
self.step_size = step_size
self.name = name
self.scale_fn = lambda x: 1 / (2. ** (x - 1))
self.clr_iterations = 0.
self.trn_iterations = 0.
def __call__(self, step):
with tf.name_scope(self.name or "CyclicalLearningRate"):
base_lr = tf.convert_to_tensor(
self.base_lr, name="base_lr"
)
dtype = base_lr.dtype
max_lr = tf.cast(self.max_lr, dtype)
step_size = tf.cast(self.step_size, dtype)
step_as_dtype = tf.cast(step, dtype)
cycle = tf.floor(1 + step_as_dtype / (2 * step_size))
x = tf.abs(step_as_dtype / step_size - 2 * cycle + 1)
return base_lr + (max_lr - base_lr) * tf.maximum(tf.cast(0, dtype), (1 - x)) * self.scale_fn(cycle)
def get_config(self):
return {
"initial_learning_rate": self.base_lr,
"maximal_learning_rate": self.max_lr,
"step_size": self.step_size,
"name": self.name
}