val_accs_raw.append(
(y_pred_raw.argmax(1) == y).float().mean().item())
val_accs_crop.append(
(y_pred_crop.argmax(1) == y).float().mean().item())
if CALLBACK:
CALLBACK.scalar('Val_Loss', epoch + 1, val_losses.last_avg)
CALLBACK.scalar('Val_Acc', epoch + 1, val_accs.last_avg)
CALLBACK.scalar('Val_Acc_Raw', epoch + 1, val_accs_raw.last_avg)
CALLBACK.scalar('Val_Acc_Crop', epoch + 1, val_accs_crop.last_avg)
print(
'\nEpoch {} | Validation | Loss: {:.4f}, Accs: {:.4f}, Accs_raw: {:.4f}, Accs_crop: {:.4f}'
.format(epoch, val_losses.avg, val_accs.avg, val_accs_raw.avg,
val_accs_crop.avg))
valacc_final = val_accs.avg
if epoch % 5 == 0 or batch_loss.item() < 0.01:
save_checkpoint(model.state_dict(),
os.path.join(
args.save_folder, args.save_name + '_valacc' +
str(valacc_final) + '_' + str(epoch) + '.pth'),
cpu=True)
print('\nSaving model...')
save_checkpoint(model.state_dict(),
os.path.join(
args.save_folder,
args.save_name + '_valacc' + str(valacc_final) + '.pth'),
cpu=True)
CALLBACK.save([args.visdom_env])
send_email("cifar10_wsdan 0.2实验完成\nval acc = " + str(valacc_final))
val_accs.append((logits.argmax(1) == labels).float().mean().item())
val_losses.append(loss.item())
if CALLBACK:
CALLBACK.scalar('Val_Loss', epoch + 1, val_losses.last_avg)
CALLBACK.scalar('Val_Acc', epoch + 1, val_accs.last_avg)
print('Epoch {} | Validation | Loss: {:.4f}, Accs: {:.4f}'.format(
epoch, val_losses.avg, val_accs.avg))
save_path = 'weights/best/'
if val_accs.avg >= best_acc: # args.adv is None and
best_acc = val_accs.avg
best_epoch = epoch
best_dict = deepcopy(model.state_dict())
files2remove = glob.glob(os.path.join(save_path, 'shape_ALP_*'))
for _i in files2remove:
os.remove(_i)
strsave = "shape_ALP_cifar10_ep_%d_val_acc%.4f.pth" % (epoch, best_acc)
torch.save(model.cpu().state_dict(), os.path.join(save_path, strsave))
model.to(DEVICE)
if args.adv is None and val_accs.avg >= best_acc:
best_acc = val_accs.avg
best_epoch = epoch
best_dict = deepcopy(model.state_dict())
if not (epoch + 1) % args.save_freq:
save_checkpoint(model.state_dict(),
os.path.join(
logits = model(images)
loss = F.cross_entropy(logits, labels)
val_accs.append((logits.argmax(1) == labels).float().mean().item())
val_losses.append(loss.item())
if CALLBACK:
CALLBACK.scalar('Val_Loss', epoch + 1, val_losses.last_avg)
CALLBACK.scalar('Val_Acc', epoch + 1, val_accs.last_avg)
print('Epoch {} | Validation | Loss: {:.4f}, Accs: {:.4f}'.format(epoch, val_losses.avg, val_accs.avg))
if args.adv is None and val_accs.avg >= best_acc:
best_acc = val_accs.avg
best_epoch = epoch
best_dict = deepcopy(model.state_dict())
if not (epoch + 1) % args.save_freq:
save_checkpoint(
model.state_dict(), os.path.join(args.save_folder, args.save_name + '_{}.pth'.format(epoch + 1)), cpu=True)
if args.adv is None:
model.load_state_dict(best_dict)
test_accs = AverageMeter()
test_losses = AverageMeter()
with torch.no_grad():
for i, (images, labels) in enumerate(tqdm.tqdm(test_loader, ncols=80)):
images, labels = images.to(DEVICE), labels.to(DEVICE)