-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathfen.py
139 lines (114 loc) · 5.47 KB
/
fen.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
'''
Suggestion for the FEN - feature extraction network.
The architecture is heavily inspired by the ResNet-18 architecture.
'''
import torch.nn as nn
import torch.nn.functional as F
def build_entry_flow():
net = nn.Sequential(nn.Conv1d(1, 3, 1),
nn.ReLU())
return net
class BasicBlock(nn.Module):
expansion = 1
def __init__(self, in_planes, planes, stride=1):
super(BasicBlock, self).__init__()
self.conv1 = nn.Conv1d(in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm1d(planes)
self.conv2 = nn.Conv1d(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
self.bn2 = nn.BatchNorm1d(planes)
self.shortcut = nn.Sequential()
if stride != 1 or in_planes != self.expansion * planes:
self.shortcut = nn.Sequential(
nn.Conv1d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm1d(self.expansion * planes)
)
def forward(self, x):
out = F.relu(self.bn1(self.conv1(x)))
out = self.bn2(self.conv2(out))
out += self.shortcut(x)
out = F.relu(out)
return out
class ResNet_1D(nn.Module):
def __init__(self, channels, levels):
super().__init__()
self.ch = channels
self.res_lev = levels
# input layer and preprocessing
self.conv1 = nn.Conv1d(1, self.ch[0], kernel_size=3, stride=1, padding=1, bias=False)
self.bn1 = nn.BatchNorm1d(self.ch[0])
self.relu = nn.LeakyReLU()
self.preprocess_level = nn.Sequential(self.conv1, self.bn1, self.relu)
self.layers = {}
in_planes = self.ch[0]
# hidden layers build:
for i in range(self.res_lev):
if i == 0:
self.layers['layer'+ str(i)] = self._make_layer(BasicBlock, in_planes, self.ch[i],
num_blocks=2, stride=1)
else:
self.layers['layer' + str(i)] = self._make_layer(BasicBlock, in_planes, self.ch[i],
num_blocks=2, stride=2)
in_planes = self.ch[i]
self.conv1x1 = nn.Conv1d(self.ch[-1], 1, kernel_size=1, stride=1, padding=1, bias=False)
self.output_level = nn.Sequential(self.conv1x1, self.relu)
self.resolution_levels = nn.ModuleList([self.layers['layer'+str(i)] for i in range(self.res_lev)])
def _make_layer(self, block, in_planes, planes, num_blocks, stride):
layers = [block(in_planes, planes, stride)]
for i in range(num_blocks - 1):
layers.append(block(planes, planes, 1))
return nn.Sequential(*layers)
def forward(self, x):
outputs = [x]
for i, layer in enumerate(self.resolution_levels):
if i == 0:
x = self.preprocess_level(x)
outputs.append(layer(x))
elif i == len(self.resolution_levels)-1:
out = layer(outputs[-1])
outputs.append(self.output_level(out))
else:
outputs.append(layer(outputs[-1]))
return outputs[-1]
class ResNet_1D_dense_output(nn.Module):
def __init__(self, channels, levels):
super().__init__()
self.ch = channels # eval(args['model']['cond_net_channels'])
self.res_lev = levels # len(eval(args['model']['inn_coupling_blocks']))
# input layer and preprocessing
self.conv1 = nn.Conv1d(1, self.ch[0], kernel_size=3, stride=1, padding=1, bias=False)
self.bn1 = nn.BatchNorm1d(self.ch[0])
self.relu = nn.LeakyReLU()
self.preprocess_level = nn.Sequential(self.conv1, self.bn1, self.relu)
self.layers = {}
in_planes = self.ch[0]
# hidden layers build:
for i in range(self.res_lev):
if i == 0:
self.layers['layer'+ str(i)] = self._make_layer(BasicBlock, in_planes, self.ch[i],
num_blocks=2, stride=1)
else:
self.layers['layer' + str(i)] = self._make_layer(BasicBlock, in_planes, self.ch[i],
num_blocks=2, stride=2)
in_planes = self.ch[i]
# make resolution layers automatically depending on default/conf.ini
self.conv1x1 = nn.Conv1d(self.ch[-1], 1, kernel_size=1, stride=1, padding=1, bias=False)
self.out_dense = nn.Linear(52, 15)
self.output_level = nn.Sequential(self.conv1x1, self.relu, self.out_dense, self.relu)
self.resolution_levels = nn.ModuleList([self.layers['layer'+str(i)] for i in range(self.res_lev)])
def _make_layer(self, block, in_planes, planes, num_blocks, stride):
layers = [block(in_planes, planes, stride)]
for i in range(num_blocks - 1):
layers.append(block(planes, planes, 1))
return nn.Sequential(*layers)
def forward(self, x):
outputs = [x]
for i, layer in enumerate(self.resolution_levels):
if i == 0:
x = self.preprocess_level(x)
outputs.append(layer(x))
elif i == len(self.resolution_levels)-1:
out = layer(outputs[-1])
outputs.append(self.output_level(out))
else:
outputs.append(layer(outputs[-1]))
return outputs[-1]