def get_args_parser()

in train.py [0:0]

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

def get_args_parser():
    # Training settings
    parser = argparse.ArgumentParser(description='PASCL for OOD detection in long-tailed recognition')
    parser.add_argument('--gpu', default='2')
    parser.add_argument('--num_workers', '--cpus', type=int, default=16, help='number of threads for data loader')
    parser.add_argument('--data_root_path', '--drp', default='./data', help='data root path')
    parser.add_argument('--dataset', '--ds', default='cifar10', choices=['cifar10', 'cifar100', 'imagenet', 'waterbird'])
    parser.add_argument('--ood_aux_dataset', '--ood_ds', default='TinyImages', choices=['TinyImages', 'VOS', 'NPOS', 'CIFAR', 'Texture'])
    parser.add_argument('--id_class_number', type=int, default=1000, help='for ImageNet subset')
    parser.add_argument('--model', '--md', default='ResNet18', choices=['ResNet18', 'ResNet34', 'ResNet50'], help='which model to use')
    parser.add_argument('--imbalance_ratio', '--rho', default=0.01, type=float)
    parser.add_argument('--seed', default=None, type=int, help='random seed')
    # training params:
    parser.add_argument('--batch_size', '-b', type=int, default=256, help='input batch size for training')
    parser.add_argument('--test_batch_size', '--tb', type=int, default=1000, help='input batch size for testing')
    parser.add_argument('--epochs', '-e', type=int, default=200, help='number of epochs to train')
    parser.add_argument('--save_epochs', type=int, default=-1, help='number of epochs to save')
    parser.add_argument('--lr', type=float, default=1e-3, help='learning rate')
    parser.add_argument('--wd', type=float, default=5e-4, help='weight decay')
    parser.add_argument('--momentum', '-m', type=float, default=0.9, help='Momentum.')
    parser.add_argument('--decay_epochs', '--de', default=[60,80], nargs='+', type=int, help='milestones for multisteps lr decay')
    parser.add_argument('--opt', default='adam', choices=['sgd', 'adam'], help='which optimizer to use')
    parser.add_argument('--decay', default='cos', choices=['cos', 'multisteps'], help='which lr decay method to use')
    parser.add_argument('--Lambda', default=0.5, type=float, help='OE loss term tradeoff hyper-parameter')
    parser.add_argument('--Lambda2', default=0.1, type=float, help='Contrastive loss term tradeoff hyper-parameter')
    parser.add_argument('--T', default=0.07, type=float, help='Temperature in NT-Xent loss (contrastive loss)')
    parser.add_argument('--k', default=0.4, type=float, help='bottom-k classes are taken as tail class')
    parser.add_argument('--num_ood_samples', default=30000, type=float, help='Number of OOD samples to use.')
    # opmitization params:
    parser.add_argument('--logit_adjust', '--tau', default=0., type=float)
    parser.add_argument('--ood_metric', default='oe', choices=['oe', 'bkg_c', 'energy', 'bin_disc', 'mc_disc', 'maha',
                                                               'ada_bin_disc', 'ada_oe', 'ada_energy', 'ada_pascl', 'ada_maha'], help='OOD training metric')
    parser.add_argument('--aux_ood_loss', default='none', choices=['none', 'pascl', 'simclr'], help='Auxilliary (e.g., feature-level) OOD training loss')
    parser.add_argument('--early-stop', action='store_true', default=True, dest='early_stop', help='If true, early stop when lambda dose not change')
    parser.add_argument('--no-early-stop', action='store_false', dest='early_stop')
    parser.add_argument('--w_beta', default=1.0, type=float)
    parser.add_argument('--t_beta', default=2.0, type=float)
    # 
    parser.add_argument('--timestamp', action='store_true', help='If true, attack time stamp after exp str')
    parser.add_argument('--resume', type=str, default='', help='Resume from pre-trained models')
    parser.add_argument('--save_root_path', '--srp', default='./runs', help='data root path')
    # ddp 
    parser.add_argument('--ddp', action='store_true', help='If true, use distributed data parallel')
    parser.add_argument('--ddp_backend', '--ddpbed', default='nccl', choices=['nccl', 'gloo', 'mpi'], help='If true, use distributed data parallel')
    parser.add_argument('--num_nodes', default=1, type=int, help='Number of nodes')
    parser.add_argument('--node_id', default=0, type=int, help='Node ID')
    parser.add_argument('--dist_url', default='tcp://localhost:23456', type=str, help='url used to set up distributed training')
    args = parser.parse_args()

    assert args.k>0, "When args.k==0, it is just the OE baseline."

    if args.dataset == 'imagenet':
        # adjust learning rate:
        args.lr *= args.batch_size / 256. # linearly scaled to batch size

    return args