def parse_args()

in train.py [0:0]

• 47 lines of code
• 7 McCabe index (conditional complexity)

def parse_args():
    parser = argparse.ArgumentParser(
        description="Train a handwriting recognition model."
    )
    parser.add_argument(
        "--config", type=str, help="A json configuration file for experiment."
    )
    parser.add_argument("--disable_cuda", action="store_true", help="Disable CUDA")
    parser.add_argument(
        "--restore", action="store_true", help="Restore training from last checkpoint")
    parser.add_argument("--last_epoch", type=int, default=0, help="Epoch restoring from.")
    parser.add_argument(
        "--checkpoint_path",
        default="/tmp/",
        type=str,
        help="Checkpoint path for saving models",
    )
    parser.add_argument(
        "--world_size", default=1, type=int, help="world size for distributed training"
    )
    parser.add_argument(
        "--dist_url",
        default="tcp://localhost:23146",
        type=str,
        help="url used to set up distributed training. This should be"
        "the IP address and open port number of the master node",
    )
    parser.add_argument(
        "--dist_backend", default="nccl", type=str, help="distributed backend"
    )
    args = parser.parse_args()
    logging.basicConfig(level=logging.INFO)

    use_cpu = args.disable_cuda or not torch.cuda.is_available()
    if args.world_size > 1 and use_cpu:
        logging.fatal("CPU distributed training not supported.")
        sys.exit(1)

    logging.info("World size is : " + str(args.world_size))
    if args.restore:
        logging.info(f"Restoring model from epoch {args.last_epoch}")

    if not use_cpu and torch.cuda.device_count() < args.world_size:
        logging.fatal(
            "At least {} cuda devices required. {} found".format(
                args.world_size, torch.cuda.device_count()
            )
        )
        sys.exit(1)

    return args