model = CH_net().cuda()
model.load_state_dict(torch.load('res_net.pth'))
config = Config(model)
config.optimizer = torch.optim.Adam(config.model.parameters(), lr=0.001)
config.lr_optim = torch.optim.lr_scheduler.CosineAnnealingLR(config.optimizer,
T_max=32)
test_transform = transforms.Compose([
transforms.ToPILImage(),
transforms.ToTensor(),
])
test_x = csv_to_tensor('data/test.csv', mode='test')
test_set = ImgDataset(test_x, transform=test_transform)
test_loader = DataLoader(test_set, batch_size=config.batch_size, shuffle=False)
#testing
model.eval()
prediction = []
with torch.no_grad():
for i, data in enumerate(test_loader):
test_pred = config.model(data.cuda())
test_label = np.argmax(test_pred.cpu().data.numpy(), axis=1)
for y in test_label:
prediction.append(y)
#write to csv
with open('prediction_0724.csv', 'w') as f:
f.write('ImageId,Label\n')
for i, y in enumerate(prediction):
f.write('{},{}\n'.format(i + 1, y))
for epoch in range(config.epoch):
'''
if not change1 and epoch>=config.epoch*1/3:
config.optimizer = torch.optim.Adam(config.model.parameters(), lr=0.0003)
change1=1
if not change2 and epoch>=config.epoch*2/3:
config.optimizer = torch.optim.SGD(config.model.parameters(), lr=0.00005)
change2=1
'''
epoch_start_time = time.time()
train_loss = 0.0
train_acc = 0.0
config.model.train()
for i, data in enumerate(train_loader):
config.optimizer.zero_grad()
train_pred = config.model(data[0].cuda())
batch_loss = config.loss(train_pred, data[1].cuda())
batch_loss.backward()
config.optimizer.step()
train_acc += np.sum(
np.argmax(train_pred.cpu().data.numpy(), axis=1) ==
data[1].numpy())
train_loss += batch_loss.item()
config.lr_optim.step()
print('[%03d/%03d] %2.2f sec(s) Train Acc: %3.6f Loss: %3.6f' % \
(epoch + 1, config.epoch, time.time()-epoch_start_time, \
train_acc/train_set.__len__(), train_loss/train_set.__len__()))