def unsqueeze(input, dim):
return np.unsqueeze(input, dim)
g h = torch.FloatTensor(1,5) h h = nn.Parameter(h) h """#Squeeze ka eg:-""" vinit = [[[[[1,2,3], [5,6,7]]]]] import numpy as np v = np.array(vinit) v v.shape import numpy as np vinit = np.squeeze(vinit) vinit vinit.shape vinit = np.unsqueeze(0)(vinit) vinit vinit.shape """#Idhar tak squeeze"""
train_loss = 0
valid_loss = 0
valid_loss_list, train_loss_list = [], []
for epoch in range(max_epochs):
train_loss = 0
valid_loss = 0
for data in train_loader:
batch_size = data.size()[0]
#print (data.size())
datav = Variable(data).cuda()
l1 = random.randint(1, 100) - 1
blb = lesion_generator()
temp_img = (datav[l1, :, :, :])
temp_img = np.unsqueeze(temp_img, 0).cpu().numpy()
temp_img[blb > 0.1] = blb[blb > 0.1]
datav[l1, :, :, :] = torch.from_numpy(temp_img)
#datav[l2,:,row2:row2+5,:]=0
mean, logvar, rec_enc = G(datav)
if epoch % 5 == 0:
beta_err = beta_loss_function(rec_enc, datav, mean, logvar, beta)
else:
beta_err = beta_loss_function(rec_enc, datav, mean, logvar, 0)
err_enc = beta_err
opt_enc.zero_grad()
err_enc.backward()
opt_enc.step()
G.eval()