print(args.__dict__)
if not os.path.exists(args.checkpoint_dir) and args.to_log == 1:
os.makedirs(args.checkpoint_dir)
os.makedirs(args.log_dir)
if args.model_type == 'pacman':
model_kwargs = {'question_vocab': load_vocab(args.vocab_json)}
shared_model = NavPlannerControllerModel(**model_kwargs)
else:
exit()
shared_model.share_memory()
if args.checkpoint_path != False:
print('Loading params from checkpoint: %s' % args.checkpoint_path)
shared_model.load_state_dict(checkpoint['state'])
if args.mode == 'eval':
eval(0, args, shared_model)
elif args.mode == 'train':
train(0, args, shared_model)
else:
print(args.__dict__)
if not os.path.exists(args.checkpoint_dir) and args.to_log == 1:
os.makedirs(args.checkpoint_dir)
os.makedirs(args.log_dir)
if args.model_type == 'pacman':
model_kwargs = {'question_vocab': load_vocab(args.vocab_json)}
shared_nav_model = NavPlannerControllerModel(**model_kwargs)
else:
exit()
shared_nav_model.share_memory()
print('Loading navigation params from checkpoint: %s' %
args.nav_checkpoint_path)
shared_nav_model.load_state_dict(checkpoint['state'])
# Load answering model
print('Loading answering checkpoint from %s' % args.ans_checkpoint_path)
ans_checkpoint = torch.load(args.ans_checkpoint_path,
map_location={'cuda:0': 'cpu'})
ans_model_kwargs = {'vocab': load_vocab(args.vocab_json)}
shared_ans_model = VqaLstmCnnAttentionModel(**ans_model_kwargs)
shared_ans_model.share_memory()
print(args.__dict__)
if not os.path.exists(args.checkpoint_dir) and args.to_log == 1:
os.makedirs(args.checkpoint_dir)
os.makedirs(args.log_dir)
if args.model_type == 'pacman':
model_kwargs = {'question_vocab': load_vocab(args.vocab_json)}
shared_nav_model = NavPlannerControllerModel(**model_kwargs)
else:
exit()
shared_nav_model.share_memory()
print('Loading navigation params from checkpoint: %s' %
args.nav_checkpoint_path)
shared_nav_model.load_state_dict(checkpoint['state'])
# Load answering model
print('Loading answering checkpoint from %s' % args.ans_checkpoint_path)
ans_checkpoint = torch.load(
args.ans_checkpoint_path, map_location={
'cuda:0': 'cpu'
})
ans_model_kwargs = {'vocab': load_vocab(args.vocab_json)}
shared_ans_model = VqaLstmCnnAttentionModel(**ans_model_kwargs)