arguments.py

import argparse

import os
import torch

dir_path = os.path.dirname(os.path.realpath(__file__))


def parse_args():
    parser = argparse.ArgumentParser(
        description="Parameters for Contrastive Predictive Coding for Human "
        "Activity Recognition"
    )

    # Data loading parameters
    parser.add_argument("--window", type=int, default=50, help="Window size")
    parser.add_argument(
        "--overlap", type=int, default=25, help="Overlap between consecutive windows"
    )

    # Training settings
    parser.add_argument("-b", "--batch_size", type=int, default=64)
    parser.add_argument("--learning_rate", type=float, default=5e-4)
    parser.add_argument("--num_epochs", type=int, default=150)
    parser.add_argument("--gpu_id", type=str, default="0")

    # Dataset to train on
    parser.add_argument(
        "--dataset",
        type=str,
        default="mobiact",
        help="Choosing the dataset to perform the training on",
    )

    # Conv encoder
    parser.add_argument(
        "--kernel_size",
        type=int,
        default=3,
        help="Size of the conv filters in the encoder",
    )

    # Future prediction horizon
    parser.add_argument(
        "--num_steps_prediction",
        type=int,
        default=28,
        help="Number of steps in the future to predict",
    )

    # ------------------------------------------------------------
    # Classification parameters
    parser.add_argument(
        "--classifier_lr",
        type=float,
        default=5e-4,
    )
    parser.add_argument("--classifier_batch_size", type=int, default=256)
    parser.add_argument(
        "--saved_model",
        type=str,
        default=None,
        help="Full path of the learned CPC model",
    )
    parser.add_argument(
        "--learning_schedule",
        type=str,
        default="last_layer",
        choices=["last_layer", "all_layers"],
        help="last layer freezes the encoder weights but " "all_layers does not.",
    )
    # ------------------------------------------------------------

    # Random seed for reproducibility
    parser.add_argument("--random_seed", type=int, default=42)

    parser.add_argument(
        "--data_percentage",
        type=int,
        default=100,
        help="Percentage of data to use for training (default: 100)",
    )

    args = parser.parse_args()

    # Setting parameters by the dataset

    args.root_dir = "../CPC/"
    args.input_size = 6

    if args.dataset == "UCI_raw_12":
        args.data_file = "UCI_raw_12"
        args.num_classes = 13
    elif args.dataset == "UCI_raw":
        args.data_file = "UCI_raw"
        args.num_classes = 7
    elif args.dataset == "MotionSense_raw":
        args.data_file = "MotionSense_raw"
        args.num_classes = 6
    elif args.dataset == "KuHar_raw":
        args.data_file = "KuHar_raw"
        args.num_classes = 18
    elif args.dataset == "RealWorld_raw":
        args.data_file = "RealWorld_raw"
        args.num_classes = 9
    elif args.dataset == "KuHar_MotionSense":
        args.data_file = "KuHar_MotionSense"
        args.num_classes = 18
    elif args.dataset == "UCI_KuHar":
        args.data_file = "UCI_KuHar"
        args.num_classes = 18
    elif args.dataset == "UCI_raw_g":
        args.data_file = "UCI_raw_g"
        args.num_classes = 7
    elif args.dataset == "UCI_MotionSense_KuHar_RealWorld":
        args.data_file = "UCI_MotionSense_KuHar_RealWorld"
        args.num_classes = 18
    elif args.dataset == "KuHar_RealWorld":
        args.data_file = "KuHar_RealWorld"
        args.num_classes = 18
    elif args.dataset == "UCI_MotionSense":
        args.data_file = "UCI_MotionSense"
        args.num_classes = 13
    elif args.dataset == "UCI_RealWorld":
        args.data_file = "UCI_RealWorld"
        args.num_classes = 9
    elif args.dataset == "MotionSense_RealWorld":
        args.data_file = "MotionSense_RealWorld"
        args.num_classes = 9
    elif args.dataset == "KuHar_UCI_RealWorld":
        args.data_file = "KuHar_UCI_RealWorld"
        args.num_classes = 18
    elif args.dataset == "KuHar_MotionSense_RealWorld":
        args.data_file = "KuHar_MotionSense_RealWorld"
        args.num_classes = 18
    elif args.dataset == "UCI_MotionSense_RealWorld":
        args.data_file = "UCI_MotionSense_RealWorld"
        args.num_classes = 13

    args.device = torch.device(
        "cuda:" + str(args.gpu_id) if torch.cuda.is_available() else "cpu"
    )

    # Conv padding size
    args.padding = int(args.kernel_size // 2)

    return args