Create feedforward_net.py

Author: gksoriginals

feedforward_net.py

@@ -0,0 +1,42 @@
+import torch
+import torch.nn as nn
+import torch.optim as optim
+from torch.utils.tensorboard import SummaryWriter
+writer = SummaryWriter('runs/dummy_model')
+x = torch.tensor([[0,0,1],[0,1,1],[1,0,1],[1,1,1]]).float()
+y = torch.tensor([[0], [1], [1], [0]]).float()
+class Net(nn.Module):
+    def __init__(self, inp, out):
+        super(Net, self).__init__()
+        self.input = nn.Linear(inp, 4)
+        self.sigmoid = nn.Sigmoid()
+        self.h1 = nn.Linear(4, 8)
+        self.h2 = nn.Linear(8, 16)
+        self.output = nn.Linear(16, out)
+
+    def forward(self, x):
+        x = self.input(x)
+        x = self.sigmoid(self.h1(x))
+        x = self.h2(x)
+        x = self.output(x)
+        return x
+
+
+model = Net(inp=3, out=1)
+for name, param in model.named_parameters():
+    if param.requires_grad:
+        print(name)
+#print(model.forward(x))
+#print(model.forward(torch.tensor([[0, 0, 1]]).float()))
+model.zero_grad()
+criterion = nn.MSELoss()
+optimr = optim.SGD(model.parameters(), lr=0.001)
+for i in range(60000):
+    output = model(x)
+    target_ = y
+    loss = criterion(output, target_)
+    print(loss)
+    loss.backward()
+    optimr.step()
+    
+print(model(x))