apm.reset()
print('val-{} Loss: {:.4f} Acc: {:.4f}'.format(dataloader.root, epoch_loss,
error))
return full_probs, epoch_loss
if __name__ == '__main__':
if args.mode == 'flow':
dataloaders, datasets = load_data(train_split, test_split, flow_root)
elif args.mode == 'rgb':
dataloaders, datasets = load_data(train_split, test_split, rgb_root)
if args.train:
model = super_event.get_super_event_model(0, classes)
criterion = nn.NLLLoss(reduce=False)
lr = 0.1 * batch_size / len(datasets['train'])
print(lr)
optimizer = optim.Adam(model.parameters(), lr=lr)
lr_sched = optim.lr_scheduler.ReduceLROnPlateau(optimizer,
patience=3,
verbose=True)
run([(model, 0, dataloaders, optimizer, lr_sched, args.model_file)],
criterion,
num_epochs=40)
else:
print('Evaluating...')
apm.reset()
print 'val-{} Loss: {:.4f} Acc: {:.4f}'.format(dataloader.root, epoch_loss,
error)
return full_probs, epoch_loss
if __name__ == '__main__':
if args.mode == 'flow':
dataloaders, datasets = load_data(train_split, test_split, flow_root)
elif args.mode == 'rgb':
dataloaders, datasets = load_data(train_split, test_split, rgb_root)
if args.train:
model = super_event.get_super_event_model(0, classes)
criterion = nn.NLLLoss(reduce=False)
lr = 0.1 * batch_size / len(datasets['train'])
print lr
optimizer = optim.Adam(model.parameters(), lr=lr)
lr_sched = optim.lr_scheduler.ReduceLROnPlateau(optimizer,
patience=3,
verbose=True)
run([(model, 0, dataloaders, optimizer, lr_sched, args.model_file)],
criterion,
num_epochs=40)
else:
print 'Evaluating...'