class ModelTrainer:
def __init__(self,
model,
optimizer,
train_loader,
test_loader,
statspath,
scheduler=None,
batch_scheduler=False,
L1lambda=0):
self.model = model
self.scheduler = scheduler
self.batch_scheduler = batch_scheduler
self.optimizer = optimizer
self.stats = ModelStats(model, statspath)
self.train = Train(
model, train_loader, optimizer, self.stats, self.scheduler
if self.scheduler and self.batch_scheduler else None, L1lambda)
self.test = Test(model, test_loader, self.stats)
def run(self, epochs=10):
pbar = tqdm_notebook(range(1, epochs + 1), desc="Epochs")
for epoch in pbar:
self.train.run()
self.test.run()
self.stats.next_epoch(
self.scheduler.get_last_lr()[0] if self.scheduler else 0)
pbar.write(self.stats.get_epoch_desc())
if self.scheduler and not self.batch_scheduler:
self.scheduler.step()
if self.scheduler:
pbar.write(
f"Learning Rate = {self.scheduler.get_last_lr()[0]:0.6f}")
# save stats for later lookup
self.stats.save()
class ModelTrainer:
def __init__(self,
statsmanager,
model,
optimizer,
train_loader,
test_loader,
statspath,
scheduler=None,
batch_scheduler=False,
L1lambda=0):
self.model = model
self.statsmanager = statsmanager
self.scheduler = scheduler
self.batch_scheduler = batch_scheduler
self.optimizer = optimizer
self.stats = ModelStats(model, statspath)
self.train = Train(statsmanager, model, train_loader, optimizer,
self.stats,
self.scheduler if self.batch_scheduler else None,
L1lambda)
self.test = Test(model, test_loader, self.stats, statsmanager,
self.scheduler)
def run(self, epochs=10):
pbar = tqdm_notebook(range(1, epochs + 1), desc="Epochs")
for epoch in pbar:
self.train.run()
self.test.run()
lr = self.optimizer.param_groups[0]['lr']
self.stats.next_epoch(lr)
pbar.write(self.stats.get_epoch_desc())
#self.statsmanager.append_lr(lr)
self.statsmanager.append_train_loss(self.stats.avg_train_loss[-1])
self.statsmanager.append_test_loss(self.stats.avg_test_loss[-1])
self.statsmanager.append_test_accuracy(100 *
self.stats.test_acc[-1])
self.statsmanager.append_train_accuracy(100 *
self.stats.train_acc[-1])
if len(self.statsmanager.data['lr']) == 0:
self.statsmanager.append_lr(self.stats.batch_lr[0])
else:
self.statsmanager.append_lr(self.stats.lr[-1])
# need to ake it more readable and allow for other schedulers
if self.scheduler and not self.batch_scheduler and not isinstance(
self.scheduler,
torch.optim.lr_scheduler.ReduceLROnPlateau):
self.scheduler.step()
pbar.write(f"Learning Rate = {lr:0.6f}")
# save stats for later lookup
self.stats.save()
class ModelTrainer:
def __init__(self,
model,
optimizer,
train_loader,
test_loader,
statspath,
criterion,
writer,
scheduler=None,
batch_scheduler=False,
L1lambda=0):
self.model = model
self.scheduler = scheduler
self.criterion = criterion
self.batch_scheduler = batch_scheduler
self.optimizer = optimizer
self.stats = ModelStats(model, statspath)
self.train = Train(model, train_loader, optimizer, self.stats,
self.scheduler if self.batch_scheduler else None,
L1lambda, criterion)
self.test = Test(model, test_loader, self.stats, writer,
self.scheduler, criterion)
self.misclass = Misclass(model, test_loader, self.stats)
self.test_loader = test_loader
torch.backends.cudnn.benchmark = True
def run(self, epochs=10):
pbar = tqdm_notebook(range(1, epochs + 1), desc="Epochs")
for epoch in pbar:
self.train.run()
self.test.run()
lr = self.optimizer.param_groups[0]['lr']
self.stats.next_epoch(lr)
pbar.write(self.stats.get_epoch_desc())
# need to ake it more readable and allow for other schedulers
if self.scheduler and not self.batch_scheduler and not isinstance(
self.scheduler,
torch.optim.lr_scheduler.ReduceLROnPlateau):
self.scheduler.step()
pbar.write(f"Learning Rate = {lr:0.6f}")
print("printing results")
printing_results(self.model, self.test_loader, epoch)
# save stats for later lookup
self.stats.save()