def train_step(model, train_loader, criterion, optimizer, epoch, step, valid_loader = None, assignment = None, loss_vector = None, loss_count=None):
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
stdev = AverageMeter()
# switch to train mode
model.train()
end = time.time()
for i, (input, target, inst_indices) in enumerate(train_loader):
# measure data loading time
data_time.update(time.time() - end)
input_var = torch.autograd.Variable(input)
# compute output
output = model(input_var)
loss, c_loss_vector, c_loss_count, c_stdev = criterion(output, list(inst_indices.numpy()), assignment, loss_vector, loss_count, input, model)
stdev.update(c_stdev, 1)
loss_vector *= 0.9
loss_count *= 0.9
loss_vector += (c_loss_vector/100.)
loss_count += c_loss_count
# measure accuracy and record loss
# prec1, prec5 = accuracy(output.data, target, topk=(1, 5))
losses.update(loss.data[0], input.size(0))
# top1.update(prec1[0], input.size(0))
# top5.update(prec5[0], input.size(0))
# compute gradient and do SGD step
model.zero_grad()
loss.backward()
optimizer.step()
# measure elapsed time
del loss, output
batch_time.update(time.time() - end)
if i % args.log_rate == 0 and i > 0:
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
# 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
# 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})\t'
.format(
epoch, i, len(train_loader), batch_time=batch_time,
data_time=data_time, loss=losses, top1=top1, top5=top5))
# Log items
logger.add_scalar("Misc/batch time (s)", batch_time.avg, step + i)
logger.add_scalar("Misc/Train_%", 1.0 - (data_time.avg/batch_time.avg), step + i)
logger.add_scalar("Misc/epoch time (min)", batch_time.sum / 60., step + i)
logger.add_scalar("Misc/time left (min)", ((len(train_loader) - i) * batch_time.avg) / 60., step + i)
logger.add_scalar(args.dataset + "/Loss train (avg)", losses.avg, step + i)
# logger.add_scalar(args.dataset + "/Perc5 train (avg)", top5.avg, step + i)
# logger.add_scalar(args.dataset + "/Perc1 train (avg)", top1.avg, step + i)
end = time.time()
if valid_loader != None and i % args.eval_step == 0 and i > 0:
_, _ = eval_step( model = model,
data_loader = valid_loader,
criterion = criterion,
step = step + i,
datasplit = "valid")
model.train()
return loss_vector, loss_count, stdev.avg
def train_step(model, train_loader, criterion, optimizer, epoch, step, valid_loader = None):
batch_time = AverageMeter()
data_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to train mode
model.train()
end = time.time()
for i, (input, target) in enumerate(train_loader):
# measure data loading time
data_time.update(time.time() - end)
target = target.cuda(async=True)
# input = input
input_var = torch.autograd.Variable(input)
target_var = torch.autograd.Variable(target)
# compute output
output = model(input_var)
loss = criterion(output, target_var)
# loss = F.nll_loss(output, target_var)
# measure accuracy and record loss
losses.update(loss.data[0], input.size(0))
# compute gradient and do SGD step
model.zero_grad()
loss.backward()
optimizer.step()
# measure elapsed time
del loss, output
batch_time.update(time.time() - end)
if i % args.log_rate == 0 and i > 0:
# print('Epoch: [{0}][{1}/{2}]\t'
# 'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
# 'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
# 'Loss {loss.val:.4f} ({loss.avg:.4f})'.format(
# epoch, i, len(train_loader), batch_time=batch_time,
# data_time=data_time, loss=losses))
print('Epoch: [{0}][{1}/{2}]\t'
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t'
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t'
'Loss {loss.val:.4f} ({loss.avg:.4f})\t'
# 'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t'
# 'Prec@5 {top5.val:.3f} ({top5.avg:.3f})'
.format(
epoch, i, len(train_loader), batch_time=batch_time,
data_time=data_time, loss=losses, top1=top1, top5=top5))
# Log items
logger.add_scalar("Misc/batch time (s)", batch_time.avg, step + i)
logger.add_scalar("Misc/Train_%", 1.0 - (data_time.avg/batch_time.avg), step + i)
logger.add_scalar("Misc/epoch time (min)", batch_time.sum / 60., step + i)
logger.add_scalar("Misc/time left (min)", ((len(train_loader) - i) * batch_time.avg) / 60., step + i)
logger.add_scalar(args.dataset + "/Loss train (avg)", losses.avg, step + i)
# logger.add_scalar(args.dataset + "/Perc5 train (avg)", top5.avg, step + i)
# logger.add_scalar(args.dataset + "/Perc1 train (avg)", top1.avg, step + i)
end = time.time()