└── FCNmodel.py


/FCNmodel.py:
--------------------------------------------------------------------------------
 1 | 
 2 | class FCN_model(nn.Module):
 3 |     def __init__(self,NumClassesOut,N_time,N_Features,N_LSTM_Out=128,N_LSTM_layers = 1
 4 |                  ,Conv1_NF = 128,Conv2_NF = 256,Conv3_NF = 128,self.lstmDropP = 0.8,self.FC_DropP = 0.3):
 5 |         super(FCN_model,self).__init__()
 6 |         
 7 |         self.N_time = N_time
 8 |         self.N_Features = N_Features
 9 |         self.NumClassesOut = NumClassesOut
10 |         self.N_LSTM_Out = N_LSTM_Out
11 |         self.N_LSTM_layers = N_LSTM_layers
12 |         self.Conv1_NF = Conv1_NF
13 |         self.Conv2_NF = Conv2_NF
14 |         self.Conv3_NF = Conv3_NF
15 |         self.lstm = nn.LSTM(self.N_Features,self.N_LSTM_Out,self.N_LSTM_layers)
16 |         self.C1 = nn.Conv1d(self.N_Features,self.Conv1_NF,8)
17 |         self.C2 = nn.Conv1d(self.Conv1_NF,self.Conv2_NF,5)
18 |         self.C3 = nn.Conv1d(self.Conv2_NF,self.Conv3_NF,3)
19 |         self.BN1 = nn.BatchNorm1d(self.Conv1_NF)
20 |         self.BN2 = nn.BatchNorm1d(self.Conv2_NF)
21 |         self.BN3 = nn.BatchNorm1d(self.Conv3_NF)
22 |         self.relu = nn.ReLU()
23 |         self.lstmDrop = nn.Dropout(lstmDropP)
24 |         self.ConvDrop = nn.Dropout(FC_DropP)
25 |         self.FC = nn.Linear(self.Conv3_NF + self.N_LSTM_Out,self.NumClassesOut)
26 |     
27 |     def init_hidden(self):
28 |         
29 |         h0 = torch.zeros(self.N_LSTM_layers, self.N_time, self.N_LSTM_Out).to(device)
30 |         c0 = torch.zeros(self.N_LSTM_layers, self.N_time, self.N_LSTM_Out).to(device)
31 |         return h0,c0
32 |     
33 |     def forward(self,x):
34 |         
35 |         # input x should be in size [B,T,F] , where B = Batch size
36 |         #                                         T = Time sampels
37 |         #                                         F = features
38 |         
39 |         h0,c0 = self.init_hidden()
40 |         x1, (ht,ct) = self.lstm(x, (h0, c0))
41 |         x1 = x1[:,-1,:]
42 |         
43 |         x2 = x.transpose(2,1)
44 |         x2 = self.ConvDrop(self.relu(self.BN1(self.C1(x2))))
45 |         x2 = self.ConvDrop(self.relu(self.BN2(self.C2(x2))))
46 |         x2 = self.ConvDrop(self.relu(self.BN3(self.C3(x2))))
47 |         x2 = torch.mean(x2,2)
48 |         
49 |         x_all = torch.cat((x1,x2),dim=1)
50 |         x_out = self.FC(x_all)
51 |         return x_out
52 | 


--------------------------------------------------------------------------------