loss = loss_fn(images, outputs)
loss_train += loss.item()
optimizer.zero_grad()
loss.backward()
optimizer.step()
if epoch == 0 or epoch % 5 == 0:
l = loss_train / images.shape[0]
model.losses.append(l)
print('epoch {} loss {:.4f} {}'.format(
epoch, l,
datetime.datetime.now().strftime('%H:%M:%S')))
end = datetime.datetime.now()
diff = end - start
print('elapsed', diff)
model.apply(init_weights)
dataloader = torch.utils.data.DataLoader(grid_dataset,
batch_size=128,
shuffle=True)
training_loop(model, optimizer, loss_fn, dataloader, n_epochs)
date_time = datetime.datetime.now().strftime("%Y_%m_%d__%H_%M_%S")
torch.save(model.state_dict(), os.path.join(models_dir,
date_time + '_model.pt'))
plt.plot(model.losses)
if opt.seed == 0:
seed = random.randint(1, 10000)
random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed(args.seed)
else:
seed = opt.seed
torch.manual_seed(opt.seed)
torch.cuda.manual_seed(opt.seed)
torch.backends.cudnn.benchmark = True
dictionary = Dictionary.load_from_file('data/dictionary.pkl')
opt.ntokens = dictionary.ntoken
model = Model(opt)
model.apply(weights_init_kn)
model = nn.DataParallel(model).cuda()
train_dset = SelfCriticalDataset(opt.split, dictionary, opt)
train_loader = DataLoader(train_dset,
opt.batch_size,
shuffle=True,
num_workers=0)
opt.use_all = 1
eval_dset = SelfCriticalDataset(opt.split_test, dictionary, opt)
eval_loader = DataLoader(eval_dset,
opt.batch_size,
shuffle=False,
num_workers=0)
train(model, train_loader, eval_loader, opt)