'v_time: {:.4f}s'.format(outputs[6]))
if args.no_tensorboard is False:
tb_writer.add_scalars('Loss', {'train': outputs[0], 'val': outputs[2]}, epoch)
tb_writer.add_scalars('Accuracy', {'train': outputs[1], 'val': outputs[3]}, epoch)
tb_writer.add_scalar('lr', outputs[4], epoch)
tb_writer.add_scalars('Time', {'train': outputs[5], 'val': outputs[6]}, epoch)
loss_train.append(outputs[0])
acc_train.append(outputs[1])
loss_val.append(outputs[2])
acc_val.append(outputs[3])
if args.early_stopping > 0 and early_stopping.early_stop:
print("Early stopping.")
classifier_model.load_state_dict(early_stopping.load_checkpoint())
break
if args.early_stopping > 0:
classifier_model.load_state_dict(early_stopping.load_checkpoint())
if args.debug:
print("Optimization Finished!")
print("Total time elapsed: {:.4f}s".format(time.time() - t_total))
# Testing
(test_adj, test_fea) = sampler.get_test_set(normalization=args.normalization, cuda=args.cuda)
if torch.cuda.is_available():
test_edges = test_adj._indices().cpu().data
else:
test_edges = test_adj._indices().data
optimizer, milestones=[200, 300, 400, 500, 600, 700], gamma=0.6)
# convert to cuda
if args.cuda:
model.cuda()
# For the mix mode, lables and indexes are in cuda.
if args.cuda or args.mixmode:
labels = labels.cuda()
idx_train = idx_train.cuda()
idx_val = idx_val.cuda()
idx_test = idx_test.cuda()
if args.warm_start is not None and args.warm_start != "":
early_stopping = EarlyStopping(fname=args.warm_start, verbose=False)
print("Restore checkpoint from %s" % (early_stopping.fname))
model.load_state_dict(early_stopping.load_checkpoint())
# set early_stopping
if args.early_stopping > 0:
early_stopping = EarlyStopping(patience=args.early_stopping, verbose=False)
print("Model is saving to: %s" % (early_stopping.fname))
if args.no_tensorboard is False:
tb_writer = SummaryWriter(
comment=f"-dataset_{args.dataset}-type_{args.type}")
def get_lr(optimizer):
for param_group in optimizer.param_groups:
return param_group['lr']
