parser.add_argument('--num_train_samples',
type=int,
default=50000,
metavar='M',
help='number of training samples (default: 3000)')
parser.add_argument('--num_test_samples',
type=int,
default=10000,
metavar='M',
help='number of test samples (default: 1000)')
parser.add_argument(
'--train_log_step',
type=int,
default=100,
metavar='M',
help='Number of iterations after which to log the loss')
global args, device
args = parser.parse_args()
args.cuda = args.cuda and torch.cuda.is_available()
cfg_from_file("config/test.yaml")
if args.cuda:
device = 'cuda'
if args.gpu_devices is None:
args.gpu_devices = [0]
else:
device = 'cpu'
main()
if len(sys.argv) == 1:
parser.print_help()
sys.exit(1)
opts = parser.parse_args()
return opts
if __name__ == '__main__':
opts = parse_args()
# print('Using config:')
# pprint.pprint(cfg)
if opts.cfg_file is not None:
cfg_from_file(opts.cfg_file)
print_cfg()
if opts.test_net is None:
qdic_dir = cfg.QUERY_DIR #osp.join(cfg.DATA_DIR, cfg.IMDB_NAME, 'query_dict')
qdic = Dictionary(qdic_dir)
qdic.load()
vocab_size = qdic.size()
test_model = models.net(opts.test_split, vocab_size, opts)
test_net_path = osp.join(get_models_dir(), 'test.prototxt')
with open(test_net_path, 'w') as f:
f.write(str(test_model))
else:
test_net_path = opts.test_net
caffe.set_mode_gpu()
from solvers.ddqn_solver import DDQNSolver
from solvers.a2c_solver import A2CSolver
from solvers.async_solver import AsyncSolver
CUR = os.path.abspath(os.path.dirname(__file__))
###################################
# Name of cfg used for experiment #
###################################
cfg_name = 'a2c.yaml'
# Parse new config into default one
if cfg_name is not None:
cfg_path = join(*[CUR, 'config', cfg_name])
cfg_from_file(cfg_path)
# Instantiate the solver
if cfg.ASYNC.USE:
solver = AsyncSolver(cfg)
else:
if cfg.TYPE == 'dqn':
solver = DQNSolver(cfg)
elif cfg.TYPE == 'ddqn':
solver = DDQNSolver(cfg)
elif cfg.TYPE == 'a2c':
solver = A2CSolver(cfg)
else:
raise NotImplementedError('TODO')
# Train