def main():
args = parse_args()
# Device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# Dataset
train_loader, test_loader, classes = mnist_loader.load_dataset(
args.dataset_dir, img_show=True)
# Model
model = Net().to(device)
print(model)
# Loss
nllloss = nn.NLLLoss().to(
device) # CrossEntropyLoss = log_softmax + NLLLoss
loss_weight = 1
centerloss = CenterLoss(10, 2).to(device)
# Optimizer
dnn_optimizer = optim.SGD(model.parameters(),
lr=0.001,
momentum=0.9,
weight_decay=0.0005)
sheduler = lr_scheduler.StepLR(dnn_optimizer, 20, gamma=0.8)
center_optimizer = optim.SGD(centerloss.parameters(), lr=0.5)
print('Start training...')
for epoch in range(100):
# Update parameters.
epoch += 1
sheduler.step()
# Train and test a model.
train_acc, train_loss, feat, labels = train(device, train_loader,
model, nllloss,
loss_weight, centerloss,
dnn_optimizer,
center_optimizer)
test_acc, test_loss = test(device, test_loader, model, nllloss,
loss_weight, centerloss)
stdout_temp = 'Epoch: {:>3}, train acc: {:<8}, train loss: {:<8}, test acc: {:<8}, test loss: {:<8}'
print(
stdout_temp.format(epoch, train_acc, train_loss, test_acc,
test_loss))
# Visualize features of each class.
vis_img_path = args.vis_img_path_temp.format(str(epoch).zfill(3))
visualize(feat.data.cpu().numpy(),
labels.data.cpu().numpy(), epoch, vis_img_path)
# Save a trained model.
model_path = args.model_path_temp.format(str(epoch).zfill(3))
torch.save(model.state_dict(), model_path)
num_classes=100,
feat_dim=embedding_dim[args.net],
no_norm=False,
use_attention=False)
params = list(net.parameters()) + list(aux_loss.parameters())
else:
aux_loss = OrthogonalProjectionLoss(no_norm=False, use_attention=False)
if args.hnc:
hnc_loss = cam_loss_kd_topk()
else:
hnc_loss = None
if args.cl:
center_loss = CenterLoss(num_classes=100, feat_dim=2048, use_gpu=True)
params = list(net.parameters()) + list(center_loss.parameters())
else:
optimizer = optim.SGD(params=params,
lr=args.lr,
momentum=0.9,
weight_decay=5e-4)
train_scheduler = optim.lr_scheduler.MultiStepLR(
optimizer, milestones=settings.MILESTONES,
gamma=0.2) #learning rate decay
iter_per_epoch = len(training_loader)
warmup_scheduler = WarmUpLR(optimizer, iter_per_epoch * args.warm)
if args.resume:
if args.pth is not None:
recent_folder = args.pth
def main():
torch.manual_seed(args.seed)
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_devices
use_gpu = torch.cuda.is_available()
if args.use_cpu: use_gpu = False
if not args.evaluate:
sys.stdout = Logger(osp.join(args.save_dir, 'log_train.txt'))
else:
sys.stdout = Logger(osp.join(args.save_dir, 'log_test.txt'))
print("==========\nArgs:{}\n==========".format(args))
if use_gpu:
print("Currently using GPU {}".format(args.gpu_devices))
cudnn.benchmark = True
torch.cuda.manual_seed_all(args.seed)
else:
print("Currently using CPU (GPU is highly recommended)")
print("Initializing dataset {}".format(args.dataset))
dataset = data_manager.init_dataset(
root=args.root,
name=args.dataset,
split_id=args.split_id,
cuhk03_labeled=args.cuhk03_labeled,
cuhk03_classic_split=args.cuhk03_classic_split,
)
transform_train = T.Compose([
T.Random2DTranslation(args.height, args.width),
T.RandomHorizontalFlip(),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
transform_test = T.Compose([
T.Resize((args.height, args.width)),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
pin_memory = True if use_gpu else False
trainloader = DataLoader(
ImageDataset(dataset.train, transform=transform_train),
batch_size=args.train_batch,
shuffle=True,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=True,
)
queryloader = DataLoader(
ImageDataset(dataset.query, transform=transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=False,
)
galleryloader = DataLoader(
ImageDataset(dataset.gallery, transform=transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=False,
)
print("Initializing model: {}".format(args.arch))
model = models.init_model(name=args.arch,
num_classes=dataset.num_train_pids,
loss={'cent'})
print("Model size: {:.5f}M".format(
sum(p.numel() for p in model.parameters()) / 1000000.0))
criterion_xent = CrossEntropyLabelSmooth(
num_classes=dataset.num_train_pids, use_gpu=use_gpu)
criterion_cent = CenterLoss(num_classes=dataset.num_train_pids,
feat_dim=model.feat_dim,
use_gpu=use_gpu)
optimizer_model = torch.optim.Adam(model.parameters(),
lr=args.lr,
weight_decay=args.weight_decay)
optimizer_cent = torch.optim.SGD(criterion_cent.parameters(),
lr=args.lr_cent)
if args.stepsize > 0:
scheduler = lr_scheduler.StepLR(optimizer_model,
step_size=args.stepsize,
gamma=args.gamma)
start_epoch = args.start_epoch
if args.resume:
print("Loading checkpoint from '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
start_epoch = checkpoint['epoch']
if use_gpu:
model = nn.DataParallel(model).cuda()
if args.evaluate:
print("Evaluate only")
test(model, queryloader, galleryloader, use_gpu)
return
start_time = time.time()
train_time = 0
best_rank1 = -np.inf
best_epoch = 0
print("==> Start training")
for epoch in range(start_epoch, args.max_epoch):
start_train_time = time.time()
train(epoch, model, criterion_xent, criterion_cent, optimizer_model,
optimizer_cent, trainloader, use_gpu)
train_time += round(time.time() - start_train_time)
if args.stepsize > 0: scheduler.step()
if args.eval_step > 0 and (epoch + 1) % args.eval_step == 0 or (
epoch + 1) == args.max_epoch:
print("==> Test")
rank1 = test(model, queryloader, galleryloader, use_gpu)
is_best = rank1 > best_rank1
if is_best:
best_rank1 = rank1
best_epoch = epoch + 1
if use_gpu:
state_dict = model.module.state_dict()
else:
state_dict = model.state_dict()
save_checkpoint(
{
'state_dict': state_dict,
'rank1': rank1,
'epoch': epoch,
}, is_best,
osp.join(args.save_dir,
'checkpoint_ep' + str(epoch + 1) + '.pth.tar'))
print("==> Best Rank-1 {:.1%}, achieved at epoch {}".format(
best_rank1, best_epoch))
elapsed = round(time.time() - start_time)
elapsed = str(datetime.timedelta(seconds=elapsed))
train_time = str(datetime.timedelta(seconds=train_time))
print(
"Finished. Total elapsed time (h:m:s): {}. Training time (h:m:s): {}.".
format(elapsed, train_time))
def main():
torch.manual_seed(args.seed)
os.environ['CUDA_VISIBLE_DEVICES'] = args.gpu_devices
use_gpu = torch.cuda.is_available()
if args.use_cpu: use_gpu = False
if not args.evaluate:
sys.stdout = Logger(osp.join(args.save_dir, 'log_train.txt'))
else:
sys.stdout = Logger(osp.join(args.save_dir, 'log_test.txt'))
print("==========\nArgs:{}\n==========".format(args))
if use_gpu:
print("Currently using GPU {}".format(args.gpu_devices))
cudnn.benchmark = True
torch.cuda.manual_seed_all(args.seed)
else:
print("Currently using CPU (GPU is highly recommended)")
print("Initializing dataset {}".format(args.dataset))
dataset = data_manager.init_img_dataset(
root=args.root,
name=args.dataset,
split_id=args.split_id,
cuhk03_labeled=args.cuhk03_labeled,
cuhk03_classic_split=args.cuhk03_classic_split,
)
transform_train = T.Compose([
T.Resize((args.height, args.width)),
T.RandomHorizontalFlip(p=0.5),
T.Pad(10),
T.RandomCrop([args.height, args.width]),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
torchvision.transforms.RandomErasing(p=0.5,
scale=(0.02, 0.4),
ratio=(0.3, 3.33),
value=(0.4914, 0.4822, 0.4465))
# T.RandomErasing(probability=0.5, sh=0.4, mean=(0.4914, 0.4822, 0.4465)),
])
transform_test = T.Compose([
T.Resize((args.height, args.width)),
T.ToTensor(),
T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
pin_memory = True if use_gpu else False
trainloader = DataLoader(
ImageDataset(dataset.train, transform=transform_train),
sampler=RandomIdentitySampler2(dataset.train,
batch_size=args.train_batch,
num_instances=args.num_instances),
batch_size=args.train_batch,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=True,
)
queryloader = DataLoader(
ImageDataset(dataset.query, transform=transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=False,
)
galleryloader = DataLoader(
ImageDataset(dataset.gallery, transform=transform_test),
batch_size=args.test_batch,
shuffle=False,
num_workers=args.workers,
pin_memory=pin_memory,
drop_last=False,
)
print("Initializing model: {}".format(args.arch))
model = models.init_model(name=args.arch,
num_classes=dataset.num_train_pids,
loss={'xent', 'htri', 'cent'})
print("Model size: {:.5f}M".format(
sum(p.numel() for p in model.parameters()) / 1000000.0))
criterion_xent = CrossEntropyLabelSmooth(
num_classes=dataset.num_train_pids, use_gpu=use_gpu)
criterion_htri = TripletLoss(margin=args.margin)
criterion_cent = CenterLoss(num_classes=dataset.num_train_pids,
feat_dim=model.feat_dim,
use_gpu=use_gpu)
optimizer_model = init_optim(args.optim, model.parameters(), args.lr,
args.weight_decay)
optimizer_cent = torch.optim.SGD(criterion_cent.parameters(),
lr=args.lr_cent)
'''only the optimizer_model use learning rate schedule'''
# if args.stepsize > 0:
# scheduler = lr_scheduler.StepLR(optimizer_model, step_size=args.stepsize, gamma=args.gamma)
'''------Modify lr_schedule here------'''
current_schedule = init_lr_schedule(schedule=args.schedule,
warm_up_epoch=args.warm_up_epoch,
half_cos_period=args.half_cos_period,
lr_milestone=args.lr_milestone,
gamma=args.gamma,
stepsize=args.stepsize)
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer_model,
lr_lambda=current_schedule)
'''------Please refer to the args.xxx for details of hyperparams------'''
#embed()
start_epoch = args.start_epoch
if args.resume:
print("Loading checkpoint from '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
model.load_state_dict(checkpoint['state_dict'])
start_epoch = checkpoint['epoch']
if use_gpu:
model = nn.DataParallel(model).cuda()
if args.evaluate:
print("Evaluate only")
test(model, queryloader, galleryloader, use_gpu)
return
start_time = time.time()
train_time = 0
best_rank1 = -np.inf
best_epoch = 0
print("==> Start training")
for epoch in range(start_epoch, args.max_epoch):
start_train_time = time.time()
train(epoch, model, criterion_xent, criterion_htri, criterion_cent,
optimizer_model, optimizer_cent, trainloader, use_gpu)
train_time += round(time.time() - start_train_time)
if args.schedule: scheduler.step()
if (epoch + 1) > args.start_eval and args.eval_step > 0 and (
epoch + 1) % args.eval_step == 0 or (epoch +
1) == args.max_epoch:
print("==> Test")
rank1 = test(model, queryloader, galleryloader, use_gpu)
is_best = rank1 > best_rank1
if is_best:
best_rank1 = rank1
best_epoch = epoch + 1
if use_gpu:
state_dict = model.module.state_dict()
else:
state_dict = model.state_dict()
save_checkpoint(
{
'state_dict': state_dict,
'rank1': rank1,
'epoch': epoch,
}, is_best,
osp.join(args.save_dir,
'checkpoint_ep' + str(epoch + 1) + '.pth.tar'))
print("==> Best Rank-1 {:.1%}, achieved at epoch {}".format(
best_rank1, best_epoch))
elapsed = round(time.time() - start_time)
elapsed = str(datetime.timedelta(seconds=elapsed))
train_time = str(datetime.timedelta(seconds=train_time))
print(
"Finished. Total elapsed time (h:m:s): {}. Training time (h:m:s): {}.".
format(elapsed, train_time))
class Trainer(BaseTrainer):
def __init__(self, config):
super(Trainer, self).__init__(config)
self.datamanager = DataManger(config["data"])
# model
self.model = Baseline(
num_classes=self.datamanager.datasource.get_num_classes("train")
)
# summary model
summary(
self.model,
input_size=(3, 256, 128),
batch_size=config["data"]["batch_size"],
device="cpu",
)
# losses
cfg_losses = config["losses"]
self.criterion = Softmax_Triplet_loss(
num_class=self.datamanager.datasource.get_num_classes("train"),
margin=cfg_losses["margin"],
epsilon=cfg_losses["epsilon"],
use_gpu=self.use_gpu,
)
self.center_loss = CenterLoss(
num_classes=self.datamanager.datasource.get_num_classes("train"),
feature_dim=2048,
use_gpu=self.use_gpu,
)
# optimizer
cfg_optimizer = config["optimizer"]
self.optimizer = torch.optim.Adam(
self.model.parameters(),
lr=cfg_optimizer["lr"],
weight_decay=cfg_optimizer["weight_decay"],
)
self.optimizer_centerloss = torch.optim.SGD(
self.center_loss.parameters(), lr=0.5
)
# learing rate scheduler
cfg_lr_scheduler = config["lr_scheduler"]
self.lr_scheduler = WarmupMultiStepLR(
self.optimizer,
milestones=cfg_lr_scheduler["steps"],
gamma=cfg_lr_scheduler["gamma"],
warmup_factor=cfg_lr_scheduler["factor"],
warmup_iters=cfg_lr_scheduler["iters"],
warmup_method=cfg_lr_scheduler["method"],
)
# track metric
self.train_metrics = MetricTracker("loss", "accuracy")
self.valid_metrics = MetricTracker("loss", "accuracy")
# save best accuracy for function _save_checkpoint
self.best_accuracy = None
# send model to device
self.model.to(self.device)
self.scaler = GradScaler()
# resume model from last checkpoint
if config["resume"] != "":
self._resume_checkpoint(config["resume"])
def train(self):
for epoch in range(self.start_epoch, self.epochs + 1):
result = self._train_epoch(epoch)
if self.lr_scheduler is not None:
self.lr_scheduler.step()
result = self._valid_epoch(epoch)
# add scalars to tensorboard
self.writer.add_scalars(
"Loss",
{
"Train": self.train_metrics.avg("loss"),
"Val": self.valid_metrics.avg("loss"),
},
global_step=epoch,
)
self.writer.add_scalars(
"Accuracy",
{
"Train": self.train_metrics.avg("accuracy"),
"Val": self.valid_metrics.avg("accuracy"),
},
global_step=epoch,
)
# logging result to console
log = {"epoch": epoch}
log.update(result)
for key, value in log.items():
self.logger.info(" {:15s}: {}".format(str(key), value))
# save model
if (
self.best_accuracy == None
or self.best_accuracy < self.valid_metrics.avg("accuracy")
):
self.best_accuracy = self.valid_metrics.avg("accuracy")
self._save_checkpoint(epoch, save_best=True)
else:
self._save_checkpoint(epoch, save_best=False)
# save logs
self._save_logs(epoch)
def _train_epoch(self, epoch):
"""Training step"""
self.model.train()
self.train_metrics.reset()
with tqdm(total=len(self.datamanager.get_dataloader("train"))) as epoch_pbar:
epoch_pbar.set_description(f"Epoch {epoch}")
for batch_idx, (data, labels, _) in enumerate(
self.datamanager.get_dataloader("train")
):
# push data to device
data, labels = data.to(self.device), labels.to(self.device)
# zero gradient
self.optimizer.zero_grad()
self.optimizer_centerloss.zero_grad()
with autocast():
# forward batch
score, feat = self.model(data)
# calculate loss and accuracy
loss = (
self.criterion(score, feat, labels)
+ self.center_loss(feat, labels) * self.config["losses"]["beta"]
)
_, preds = torch.max(score.data, dim=1)
# backward parameters
# loss.backward()
self.scaler.scale(loss).backward()
# backward parameters for center_loss
for param in self.center_loss.parameters():
param.grad.data *= 1.0 / self.config["losses"]["beta"]
# optimize
# self.optimizer.step()
self.scaler.step(self.optimizer)
self.optimizer_centerloss.step()
self.scaler.update()
# update loss and accuracy in MetricTracker
self.train_metrics.update("loss", loss.item())
self.train_metrics.update(
"accuracy",
torch.sum(preds == labels.data).double().item() / data.size(0),
)
# update process bar
epoch_pbar.set_postfix(
{
"train_loss": self.train_metrics.avg("loss"),
"train_acc": self.train_metrics.avg("accuracy"),
}
)
epoch_pbar.update(1)
return self.train_metrics.result()
def _valid_epoch(self, epoch):
"""Validation step"""
self.model.eval()
self.valid_metrics.reset()
with torch.no_grad():
with tqdm(total=len(self.datamanager.get_dataloader("val"))) as epoch_pbar:
epoch_pbar.set_description(f"Epoch {epoch}")
for batch_idx, (data, labels, _) in enumerate(
self.datamanager.get_dataloader("val")
):
# push data to device
data, labels = data.to(self.device), labels.to(self.device)
with autocast():
# forward batch
score, feat = self.model(data)
# calculate loss and accuracy
loss = (
self.criterion(score, feat, labels)
+ self.center_loss(feat, labels)
* self.config["losses"]["beta"]
)
_, preds = torch.max(score.data, dim=1)
# update loss and accuracy in MetricTracker
self.valid_metrics.update("loss", loss.item())
self.valid_metrics.update(
"accuracy",
torch.sum(preds == labels.data).double().item() / data.size(0),
)
# update process bar
epoch_pbar.set_postfix(
{
"val_loss": self.valid_metrics.avg("loss"),
"val_acc": self.valid_metrics.avg("accuracy"),
}
)
epoch_pbar.update(1)
return self.valid_metrics.result()
def _save_checkpoint(self, epoch, save_best=True):
"""save model to file"""
state = {
"epoch": epoch,
"state_dict": self.model.state_dict(),
"center_loss": self.center_loss.state_dict(),
"optimizer": self.optimizer.state_dict(),
"optimizer_centerloss": self.optimizer_centerloss.state_dict(),
"lr_scheduler": self.lr_scheduler.state_dict(),
"best_accuracy": self.best_accuracy,
}
filename = os.path.join(self.checkpoint_dir, "model_last.pth")
self.logger.info("Saving last model: model_last.pth ...")
torch.save(state, filename)
if save_best:
filename = os.path.join(self.checkpoint_dir, "model_best.pth")
self.logger.info("Saving current best: model_best.pth ...")
torch.save(state, filename)
def _resume_checkpoint(self, resume_path):
"""Load model from checkpoint"""
if not os.path.exists(resume_path):
raise FileExistsError("Resume path not exist!")
self.logger.info("Loading checkpoint: {} ...".format(resume_path))
checkpoint = torch.load(resume_path, map_location=self.map_location)
self.start_epoch = checkpoint["epoch"] + 1
self.model.load_state_dict(checkpoint["state_dict"])
self.center_loss.load_state_dict(checkpoint["center_loss"])
self.optimizer.load_state_dict(checkpoint["optimizer"])
self.optimizer_centerloss.load_state_dict(checkpoint["optimizer_centerloss"])
self.lr_scheduler.load_state_dict(checkpoint["lr_scheduler"])
self.best_accuracy = checkpoint["best_accuracy"]
self.logger.info(
"Checkpoint loaded. Resume training from epoch {}".format(self.start_epoch)
)
def _save_logs(self, epoch):
"""Save logs from google colab to google drive"""
if os.path.isdir(self.logs_dir_saved):
shutil.rmtree(self.logs_dir_saved)
destination = shutil.copytree(self.logs_dir, self.logs_dir_saved)
class Trainer(object):
#cuda = torch.cuda.is_available()
#torch.backends.cudnn.benchmark = True
def __init__(self, model, optimizer, loss_f, save_dir=None, save_freq=1):
self.model = model
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if torch.cuda.device_count() > 1:
print("Let's use", torch.cuda.device_count(), "GPUs!")
# dim = 0 [30, xxx] -> [10, ...], [10, ...], [10, ...] on 3 GPUs
self.model = torch.nn.DataParallel(self.model,
device_ids=range(
torch.cuda.device_count()))
self.model.to(device)
#self.model.load_state_dict(torch.load("checkpoints/exp1/ model_370.pkl")['weight'])
self.optimizer = optimizer
self.loss_1 = loss_f().cuda()
self.loss_2 = CenterLoss().cuda()
self.optimizer_center = torch.optim.Adam(
params=self.loss_2.parameters())
self.save_dir = save_dir
self.save_freq = save_freq
self.writer = SummaryWriter()
def _iteration(self, data_loader, ep, is_train=True):
loop_loss = []
#loop_loss_a = []
outputlabel = []
targetlabel = []
for img1, target in tqdm(data_loader):
img1, target = img1.cuda(), target.cuda()
target = target.squeeze_()
out1, out2 = self.model(img1)
print(out2.size())
loss_1 = self.loss_1(out2, target)
loss_2 = self.loss_2(out1, target)
loss_step = loss_1.data.item()
print(">>>loss:", loss_step)
loop_loss.append(loss_1.data.item() / len(data_loader))
#loop_loss_a.append(loss_a.data.item() / len(data_loader))
#accuracy.append((output.data.max(1)[1] == target.data).sum().item())
if is_train:
self.optimizer.zero_grad()
self.optimizer_center.zero_grad()
loss_1.backward(retain_graph=True)
loss_2.backward(retain_graph=True)
self.optimizer_center.step()
self.optimizer.step()
output = F.softmax(out2, dim=1)
output = output.cpu()
output = output.data.numpy()
output = np.argmax(output, axis=1)
target = target.cpu().data.numpy()
# target = np.argmax(target,axis=1)
target = np.reshape(target, [-1])
output = np.reshape(output, [-1])
target = target.astype(np.int8)
output = output.astype(np.int8)
outputlabel.append(output)
targetlabel.append(target)
if is_train:
self.writer.add_scalar('train/loss_epoch', sum(loop_loss), ep)
targetlabel = np.reshape(np.array(targetlabel),
[-1]).astype(np.int)
outputlabel = np.reshape(np.array(outputlabel),
[-1]).astype(np.int)
accuracy = accuracy_score(targetlabel, outputlabel)
self.writer.add_scalar('train/accuracy', accuracy, ep)
#self.writer.add_scalar('train/loss_a_epoch', sum(loop_loss_a), ep)
#self.writer.add_scalar('train/accuracy',sum(accuracy)/len(data_loader.dataset),ep)
else:
print(targetlabel)
print(outputlabel)
targetlabel = np.reshape(np.array(targetlabel),
[-1]).astype(np.int)
outputlabel = np.reshape(np.array(outputlabel),
[-1]).astype(np.int)
accuracy = accuracy_score(targetlabel, outputlabel)
print(accuracy)
matrixs = confusion_matrix(targetlabel, outputlabel)
np.save('matrixs/matrixs_' + str(ep) + '.npy', matrixs)
self.writer.add_scalar('test/accuracy', accuracy, ep)
self.writer.add_scalar('test/loss_epoch', sum(loop_loss), ep)
#self.writer.add_scalar('test/loss_a_epoch', sum(loop_loss_a), ep)
#self.writer.add_scalar('test/accuracy',sum(accuracy)/len(data_loader.dataset),ep)
mode = "train" if is_train else "test"
#print(">>>[{mode}] loss: {loss}/accuracy: {accuracy}".format(mode=mode,loss=sum(loop_loss),accuracy=sum(accuracy)/len(data_loader.dataset)))
print(">>>[{mode}] loss: {loss}".format(mode=mode,
loss=sum(loop_loss)))
return loop_loss
def train(self, data_loader, ep):
self.model.train()
with torch.enable_grad():
loss = self._iteration(data_loader, ep)
#pass
def test(self, data_loader, ep):
self.model.eval()
with torch.no_grad():
loss = self._iteration(data_loader, ep, is_train=False)
def loop(self, epochs, train_data, test_data, scheduler=None):
for ep in range(1, epochs + 1):
if scheduler is not None:
scheduler.step()
print("epochs: {}".format(ep))
self.train(train_data, ep)
if (ep % self.save_freq == 0):
self.save(ep)
self.test(test_data, ep)
def save(self, epoch, **kwargs):
model_out_path = self.save_dir
state = {"epoch": epoch, "weight": self.model.state_dict()}
if not os.path.exists(model_out_path):
os.makedirs(model_out_path)
torch.save(state,
model_out_path + '/ model_{epoch}.pkl'.format(epoch=epoch))