|
a |
|
b/py_version/model.py |
|
|
1 |
import torch |
|
|
2 |
import torch.nn as nn |
|
|
3 |
import torch.nn.functional as F |
|
|
4 |
|
|
|
5 |
class ConvBlock(nn.Module): |
|
|
6 |
|
|
|
7 |
def __init__(self, in_channels, out_channels, kernel, stride): |
|
|
8 |
super(ConvBlock, self).__init__() |
|
|
9 |
|
|
|
10 |
self.layer = nn.Sequential( |
|
|
11 |
nn.Conv1d(in_channels = in_channels, out_channels = out_channels, kernel_size = kernel, stride = stride), |
|
|
12 |
nn.BatchNorm1d(out_channels), |
|
|
13 |
nn.ReLU() |
|
|
14 |
) |
|
|
15 |
|
|
|
16 |
def forward(self, x): |
|
|
17 |
|
|
|
18 |
out = self.layer(x) |
|
|
19 |
|
|
|
20 |
return out |
|
|
21 |
|
|
|
22 |
|
|
|
23 |
class CNN1D(nn.Module): |
|
|
24 |
|
|
|
25 |
def __init__(self, num_classes): |
|
|
26 |
super(CNN1D, self).__init__() |
|
|
27 |
|
|
|
28 |
self.layer = nn.Sequential( |
|
|
29 |
ConvBlock(16, 16, 10, 4), |
|
|
30 |
ConvBlock(16, 16, 5, 2), |
|
|
31 |
ConvBlock(16, 16, 5, 2), |
|
|
32 |
ConvBlock(16, 32, 5, 2), |
|
|
33 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
34 |
ConvBlock(32, 32, 4, 1), |
|
|
35 |
ConvBlock(32, 32, 4, 1), |
|
|
36 |
ConvBlock(32, 64, 4, 1), |
|
|
37 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
38 |
ConvBlock(64, 64, 3, 1), |
|
|
39 |
ConvBlock(64, 64, 3, 1), |
|
|
40 |
ConvBlock(64, 128, 3, 1), |
|
|
41 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
42 |
ConvBlock(128, 128, 2, 1), |
|
|
43 |
ConvBlock(128, 128, 2, 1), |
|
|
44 |
ConvBlock(128, 256, 2, 1), |
|
|
45 |
nn.MaxPool1d(kernel_size = 2, stride = 2) |
|
|
46 |
) |
|
|
47 |
|
|
|
48 |
self.linear = nn.Sequential( |
|
|
49 |
nn.Linear(1280, 512), |
|
|
50 |
nn.ReLU(), |
|
|
51 |
nn.Dropout(0.5), |
|
|
52 |
nn.Linear(512, 128), |
|
|
53 |
nn.ReLU(), |
|
|
54 |
nn.Linear(128, num_classes) |
|
|
55 |
) |
|
|
56 |
|
|
|
57 |
def forward(self, x): |
|
|
58 |
|
|
|
59 |
batch_size = x.shape[0] |
|
|
60 |
out = self.layer(x) |
|
|
61 |
out = out.view(batch_size, -1) |
|
|
62 |
out = self.linear(out) |
|
|
63 |
|
|
|
64 |
return out |
|
|
65 |
|
|
|
66 |
|
|
|
67 |
class CNN1D_F(nn.Module): |
|
|
68 |
|
|
|
69 |
def __init__(self, num_classes): |
|
|
70 |
super(CNN1D_F, self).__init__() |
|
|
71 |
|
|
|
72 |
self.layer = nn.Sequential( |
|
|
73 |
ConvBlock(16, 16, 10, 4), |
|
|
74 |
ConvBlock(16, 16, 5, 2), |
|
|
75 |
ConvBlock(16, 16, 5, 2), |
|
|
76 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
77 |
ConvBlock(16, 16, 5, 2), |
|
|
78 |
ConvBlock(16, 16, 5, 2), |
|
|
79 |
ConvBlock(16, 32, 5, 2), |
|
|
80 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
81 |
ConvBlock(32, 32, 4, 1), |
|
|
82 |
ConvBlock(32, 32, 4, 1), |
|
|
83 |
ConvBlock(32, 64, 4, 1), |
|
|
84 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
85 |
ConvBlock(64, 64, 3, 1), |
|
|
86 |
ConvBlock(64, 64, 3, 1), |
|
|
87 |
ConvBlock(64, 128, 3, 1), |
|
|
88 |
nn.MaxPool1d(kernel_size = 2, stride = 2), |
|
|
89 |
ConvBlock(128, 128, 2, 1), |
|
|
90 |
ConvBlock(128, 128, 2, 1), |
|
|
91 |
ConvBlock(128, 256, 2, 1), |
|
|
92 |
nn.MaxPool1d(kernel_size = 2, stride = 2) |
|
|
93 |
) |
|
|
94 |
|
|
|
95 |
self.linear = nn.Sequential( |
|
|
96 |
nn.Linear(1280, 512), |
|
|
97 |
nn.ReLU(), |
|
|
98 |
nn.Dropout(0.5), |
|
|
99 |
nn.Linear(512, 128), |
|
|
100 |
nn.ReLU(), |
|
|
101 |
nn.Linear(128, num_classes) |
|
|
102 |
) |
|
|
103 |
|
|
|
104 |
def forward(self, x): |
|
|
105 |
|
|
|
106 |
batch_size = x.shape[0] |
|
|
107 |
out = self.layer(x) |
|
|
108 |
out = out.view(batch_size, -1) |
|
|
109 |
out = self.linear(out) |
|
|
110 |
|
|
|
111 |
return out |