VIT Pruning 3부 train_cifa10.py

##### Validation
import time
def test(epoch):
    global best_acc
    net.eval()
    test_loss = 0
    correct = 0
    total = 0
    with torch.no_grad():
        for batch_idx, (inputs, targets) in enumerate(testloader):
            inputs, targets = inputs.to(device), targets.to(device)
            outputs = net(inputs)
            loss = criterion(outputs, targets)

            test_loss += loss.item()
            _, predicted = outputs.max(1)
            total += targets.size(0)
            correct += predicted.eq(targets).sum().item()

            progress_bar(batch_idx, len(testloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
                % (test_loss/(batch_idx+1), 100.*correct/total, correct, total))
    
    # Update scheduler
    if not args.cos:
        scheduler.step(test_loss)
    
    # Save checkpoint.
    acc = 100.*correct/total
    if acc > best_acc:
        print('Saving..')
        state = {
            'net': net.state_dict(),
            'acc': acc,
            'epoch': epoch,
        }
        if not os.path.isdir('checkpoint'):
            os.mkdir('checkpoint')
        torch.save(state, './checkpoint/'+args.net+'-{}-ckpt.t7'.format(args.patch))
        best_acc = acc
    
    os.makedirs("log", exist_ok=True)
    content = time.ctime() + ' ' + f'Epoch {epoch}, lr: {optimizer.param_groups[0]["lr"]:.7f}, val loss: {test_loss:.5f}, acc: {(acc):.5f}'
    print(content)
    with open(f'log/log_{args.net}_patch{args.patch}.txt', 'a') as appender:
        appender.write(content + "\n")
    return test_loss, acc

with torch.no_grad():
        for batch_idx, (inputs, targets) in enumerate(testloader):
            inputs, targets = inputs.to(device), targets.to(device)
            outputs = net(inputs)
            loss = criterion(outputs, targets)

            test_loss += loss.item()
            _, predicted = outputs.max(1)
            total += targets.size(0)
            correct += predicted.eq(targets).sum().item()

            progress_bar(batch_idx, len(testloader), 'Loss: %.3f | Acc: %.3f%% (%d/%d)'
                % (test_loss/(batch_idx+1), 100.*correct/total, correct, total))

.size : 텐서 사이즈 확인 설명, dim과 다르다
.eq : 같다라는 연산 설명
progress_bar : util.py 에 있음

    # Update scheduler
    if not args.cos:
        scheduler.step(test_loss)

.step : gradient 값을 이용하여 파라미터를 업데이트 함. 설명1, 설명2

    # Save checkpoint.
    acc = 100.*correct/total
    if acc > best_acc:
        print('Saving..')
        state = {
            'net': net.state_dict(),
            'acc': acc,
            'epoch': epoch,
        }
        if not os.path.isdir('checkpoint'):
            os.mkdir('checkpoint')
        torch.save(state, './checkpoint/'+args.net+'-{}-ckpt.t7'.format(args.patch))
        best_acc = acc
    
    os.makedirs("log", exist_ok=True)
    content = time.ctime() + ' ' + f'Epoch {epoch}, lr: {optimizer.param_groups[0]["lr"]:.7f}, val loss: {test_loss:.5f}, acc: {(acc):.5f}'
    print(content)
    with open(f'log/log_{args.net}_patch{args.patch}.txt', 'a') as appender:
        appender.write(content + "\n")
    return test_loss, acc

time.ctime() : 설명
open : 설명1, 설명2

list_loss = []
list_acc = []
for epoch in range(start_epoch, args.n_epochs):
    trainloss = train(epoch)
    val_loss, acc = test(epoch)
    
    if args.cos:
        scheduler.step(epoch-1)
    
    list_loss.append(val_loss)
    list_acc.append(acc)
    
    # write as csv for analysis
    with open(f'log/log_{args.net}_patch{args.patch}.csv', 'w') as f:
        writer = csv.writer(f, lineterminator='\n')
        writer.writerow(list_loss) 
        writer.writerow(list_acc) 
    # print(list_loss)

csv.writer : 설명