features = args.nfeatures
model = models.GraphNetwork(args.model_config,
features,
multigpu=args.multigpu,
default_fnet_widths=args.fnet_widths,
default_fnet_llbias=args.fnet_llbias,
default_edge_attr=args.edge_attr,
default_conv_bias=args.conv_bias,
default_fnet_tanh=args.fnet_tanh)
print('loading pretrain')
if (os.path.isfile(args.pretrain_path)):
_, _, model_state, _ = utils.load_checkpoint(args.pretrain_path)
model.load_state_dict(model_state)
else:
print('Wrong pretrain path')
exit()
print(model)
if args.cuda is True and args.multigpu is False:
model = model.to('cuda:0')
label_path = os.path.join(args.dataset_path, args.classname)
if not os.path.isfile(label_path):
raise RuntimeError("label file does not exist")
label_names = utils.read_string_list(label_path)
assert args.batch_size % args.batch_parts == 0
args.fnet_widths = ast.literal_eval(args.fnet_widths)
features = args.nfeatures
model = models.GraphNetwork(args.model_config,
features,
multigpu=args.multigpu,
default_fnet_widths=args.fnet_widths,
default_fnet_llbias=args.fnet_llbias,
default_edge_attr=args.edge_attr,
default_conv_bias=args.conv_bias,
default_fnet_tanh=args.fnet_tanh)
print('loading pretrain')
if (os.path.exists(args.pretrain_path)):
_, stored_mean_acc, model_state, _ = utils.load_checkpoint(
args.pretrain_path)
model.load_state_dict(model_state)
else:
print('Wrong pretrain path')
exit()
if args.cuda is True and args.multigpu is False:
model = model.to('cuda:0')
print(model)
label_path = os.path.join(args.dataset_path, args.classname)
if not os.path.isfile(label_path):
raise RuntimeError("label file does not exist")
label_names = utils.read_string_list(label_path)
num_workers=args.nworkers,
shuffle=False,
pin_memory=False)
is_best_meanacc = False
best_meanacc = 0
start_epoch = 0
resume_done = 0
if args.resume:
checkpoint_path_file = os.path.join(checkpoint_path,
'checkpoint_latest.pth.tar')
if (os.path.isfile(checkpoint_path_file)):
resume_done = 1
epoch, best_meanacc, model_state, optimizer_state = utils.load_checkpoint(
checkpoint_path_file)
start_epoch = epoch + 1
model.load_state_dict(model_state)
if len(optimizer_state['state']) > 0:
optimizer.load_state_dict(optimizer_state['optimizer'])
else:
print('There are problems with the optimizer state')
else:
print('Checkpoint does not exist, starting new trainning')
if args.transfer_learning != '' and args.transfer_learning != '-' and resume_done == 0:
if not os.path.isfile(args.transfer_learning):
raise RuntimeError("Transfer learning model does not exist")
model_dict = model.state_dict()