Update 0523

Add style_transfer and optim.

optim/main.py

@@ -0,0 +1,52 @@
+"""
+Optim x^2 + xy + y^2 +4
+"""
+
+import numpy as np
+
+import torch
+import torch.nn as nn
+from torch.autograd import Variable
+import torch.optim as optim
+
+# Define my own loss function
+class Optim_Function(nn.Module):
+
+    def __init__(self):
+        super(Optim_Function, self).__init__()
+
+    def forward(self, x):
+        self.loss = torch.add(torch.add(torch.add(torch.pow(x[0], 2), torch.mul(x[0], x[1])), torch.pow(x[1], 2)), Variable(torch.Tensor([4])))
+        return self.loss
+
+    def backward(self):
+        self.loss.backward()
+
+xy = Variable(torch.Tensor([300, 400]), requires_grad=True)
+
+loss = torch.add(torch.add(torch.add(torch.pow(xy[0], 2), torch.mul(xy[0], xy[1])), torch.pow(xy[1], 2)), Variable(torch.Tensor([4])))
+
+print "loss: ", loss
+loss.backward()
+print(xy.grad)
+
+criterion = Optim_Function()
+print "res: ", criterion(xy)
+
+xy = nn.Parameter(xy.data)
+optimizer = optim.SGD([xy], lr=0.1)
+#optimizer = optim.LBFGS([xy])
+
+run = [0]
+while run[0] <= 100:
+
+    def closure():
+        optimizer.zero_grad()
+        loss = torch.add(torch.add(torch.add(torch.pow(xy[0], 2), torch.mul(xy[0], xy[1])), torch.pow(xy[1], 2)),
+                         Variable(torch.Tensor([4])))
+        loss.backward()
+        print "loss: ", loss.data[0]
+        run[0] += 1
+        return loss
+
+    optimizer.step(closure)