ADD

# Add model module:
#     Add the Net1 model interface

# Add a module group:
#     Implement the PreNet network model structure
#     Implement the HighwayNet network model structure

model/Net1.py

@@ -0,0 +1,13 @@
+from torch.nn import Module
+
+
+class Net1(Module):
+    def __init__(self):
+        super().__init__()
+
+        # TODO : init the net1 model
+
+    def forward(self):
+        # TODO : implement the net1 structure
+
+        pass

model/modules.py

@@ -0,0 +1,50 @@
+import torch
+from torch.nn import Module, init
+from torch.nn import Linear, Dropout, ReLU, Sigmoid
+
+
+class PreNet(Module):
+    def __init__(self, in_dims, out_dims_1, out_dims_2, dropout_rate):
+        super(PreNet, self).__init__()
+
+        self.relu = ReLU()
+        self.drop = Dropout(dropout_rate)
+
+        self.fc1 = Linear(in_dims, out_dims_1)
+        self.fc2 = Linear(out_dims_1, out_dims_2)
+
+    def froward(self, inputs):
+        fc1_outputs = self.fc1(inputs)
+        relu1_outputs = self.relu(fc1_outputs)
+        layer_1_outputs = self.drop(relu1_outputs)
+
+        fc2_outputs = self.fc2(layer_1_outputs)
+        relu2_outputs = self.relu(fc2_outputs)
+        layer_2_outputs = self.drop(relu2_outputs)
+
+        return layer_2_outputs
+
+
+class HighwayNet(Module):
+    def __init__(self, in_dims, out_dims):
+        super(HighwayNet, self).__init__()
+
+        self.fc1 = Linear(in_dims, out_dims)
+        self.fc2 = Linear(in_dims, out_dims)
+        init.constant_(self.fc2.bias, -1.0)
+
+        self.relu = ReLU()
+        self.sigmoid = Sigmoid()
+
+    def froward(self, inputs):
+        h = self.fc1(inputs)
+        H = self.relu(h)
+
+        t = self.fc2(inputs)
+        T = self.sigmoid(t)
+
+        C = 1.0 - T
+
+        outputs = H * T + inputs * C
+
+        return outputs