grad_scale = grad_scale / 8.0
logger("training phase")
for batch_idx, (data, target) in enumerate(train_loader):
indx_target = target.clone()
if args.cuda:
data, target = data.cuda(), target.cuda()
data, target = Variable(data), Variable(target)
optimizer.zero_grad()
output = model(data)
loss = wage_util.SSE(output, target)
loss.backward()
for name, param in list(model.named_parameters())[::-1]:
param.grad.data = wage_quantizer.QG(param.grad.data,
args.wl_grad, grad_scale)
optimizer.step()
for name, param in list(model.named_parameters())[::-1]:
param.data = wage_quantizer.C(param.data, args.wl_weight)
if batch_idx % args.log_interval == 0 and batch_idx > 0:
pred = output.data.max(1)[
1] # get the index of the max log-probability
correct = pred.cpu().eq(indx_target).sum()
acc = float(correct) * 1.0 / len(data)
logger(
'Train Epoch: {} [{}/{}] Loss: {:.6f} Acc: {:.4f} lr: {:.2e}'
.format(epoch, batch_idx * len(data),
len(train_loader.dataset), loss.data, acc,
if args.cuda:
data, target = data.cuda(), target.cuda()
data, target = Variable(data), Variable(target)
optimizer.zero_grad()
output = model(data)
loss = wage_util.SSE(output, target)
loss.backward()
# introduce non-ideal property
j = 0
for name, param in list(model.named_parameters())[::-1]:
velocity[j] = gamma * velocity[j] + alpha * param.grad.data
param.grad.data = velocity[j]
param.grad.data = wage_quantizer.QG(
param.data, args.wl_weight, param.grad.data, args.wl_grad,
grad_scale,
torch.from_numpy(paramALTP[j]).cuda(),
torch.from_numpy(paramALTD[j]).cuda(), args.max_level,
args.max_level)
j = j + 1
optimizer.step()
for name, param in list(model.named_parameters())[::-1]:
param.data = wage_quantizer.W(param.data, param.grad.data,
args.wl_weight, args.c2cVari)
if batch_idx % args.log_interval == 0 and batch_idx > 0:
pred = output.data.max(1)[
1] # get the index of the max log-probability
correct = pred.cpu().eq(indx_target).sum()
acc = float(correct) * 1.0 / len(data)