Skip to content

why use the drnet-py can't accomplish the train result compare with lua train file? #6

Description

@Feifannaner

Thanks for you python code .
when I use the lua file to train the single number moving mnist data ,the scene desciminator can accomplish the 98%+ train result. however, when i run it using the python files ,it cannot accomplish the same result.
train_issue
could you explain it ? or there is something wrong with the code? @edenton

`
def train_scene_discriminator(x):
netC.zero_grad()
target = torch.cuda.FloatTensor(opt.batch_size,1).to(device)
import random
offset1 = random.randint(0,opt.max_step-1)$
offset2 = random.randint(0,opt.max_step-1)$
x1 = x[offset1].to(device)$
x2 = x[offset2].to(device)$
#(bs,t,w,h,n)$

h_p1 = netEP(x1).detach()$
h_p2 = netEP(x2).detach()$
half = int(opt.batch_size/2)$
rp = torch.randperm(half).to(device)$
#h_p2[:half] = h_p2[rp]$
h_p2[half:] = h_p2[opt.batch_size/2+rp]$
target[:half] = 1$
target[half:] = 0$
out = netC([h_p1, h_p2])$
bce = bce_criterion(out, Variable(target))$
bce.backward()$
optimizerC.step()$

`

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions