initial commit

solver.py

@@ -76,77 +76,6 @@ class Solver(object):
         self.batch_size = args.batch_size
         self.data_loader = return_data(args)
 
-#    def show_factorization(self):
-#        decode = self.net.decode
-#        interpolation = torch.arange(-2, 2.1, 0.5)
-#        for row in range(10):
-#            samples = []
-#            for _ in range(5):
-#                z = Variable(cuda(torch.randn(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
-#                for val in interpolation:
-#                    z[:, row] = val
-#                    sample = F.sigmoid(decode(z))
-#                    samples.append(sample)
-#            samples = torch.cat(samples, dim=0).data.cpu()
-#            self.viz.images(samples, opts=dict(title='row:'+str(row)), nrow=len(interpolation))
-
-
-    # get z|x
-    def show_factorization(self):
-        decode = self.net.decode
-        #interpolation = torch.arange(-2, 2.1, 0.5)
-        interpolation = torch.arange(-6, 6.1, 2)
-        num_sample = 20
-        for i in range(num_sample):
-            img = iter(self.data_loader).next()[1]
-            img = Variable(cuda(img, self.use_cuda))
-            z_ori = self.net.encode(img)[:1, :self.z_dim]
-            z_ori = Variable(cuda(torch.randn(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
-            samples = []
-            for row in range(10):
-                z = z_ori.clone()
-                for val in interpolation:
-                    z[:, row] = val
-                    sample = F.sigmoid(decode(z))
-                    samples.append(sample)
-            samples = torch.cat(samples, dim=0).data.cpu()
-            self.viz.images(samples, opts=dict(title='sample:'+str(i)), nrow=len(interpolation))
-
-#    # fix z random
-#    def show_factorization(self):
-#        decode = self.net.decode
-#        interpolation = torch.arange(-2, 2.1, 0.5)
-#        num_sample = 20
-#        for i in range(num_sample):
-#            z_ori = Variable(cuda(torch.randn(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
-#            samples = []
-#            for row in range(10):
-#                z = z_ori.clone()
-#                for val in interpolation:
-#                    z[:, row] = val
-#                    sample = F.sigmoid(decode(z))
-#                    samples.append(sample)
-#            samples = torch.cat(samples, dim=0).data.cpu()
-#            self.viz.images(samples, opts=dict(title='sample:'+str(i)), nrow=len(interpolation))
-
-#    # fix z(zero vector)
-#    def show_factorization(self):
-#        decode = self.net.decode
-#        interpolation = torch.arange(-2, 2.1, 0.5)
-#        num_sample = 1
-#        for i in range(num_sample):
-#            #z_ori = Variable(cuda(torch.randn(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
-#            z_ori = Variable(cuda(torch.zeros(1, self.z_dim, 1, 1), self.use_cuda), volatile=True)
-#            samples = []
-#            for row in range(10):
-#                z = z_ori.clone()
-#                for val in interpolation:
-#                    z[:, row] = val
-#                    sample = F.sigmoid(decode(z))
-#                    samples.append(sample)
-#            samples = torch.cat(samples, dim=0).data.cpu()
-#            self.viz.images(samples, opts=dict(title='sample:'+str(i)), nrow=len(interpolation))
-
     def train(self):
         self.net_mode(train=True)
         out = False