main.py

from src.datasets import Circles, Tori, Disks
from src.models import Classifier1L, ClassifierAL
from src.experiments import ActivationExperiments, TopologyChangeExperiments
import argparse
import time

DATASETS_DICT = {"circles": Circles, "tori": Tori, "disks": Disks}

EXPERIMENTS_DICT = {
    "activation": ActivationExperiments,
    "topo_change": TopologyChangeExperiments,
}

MODELS_DICT = {"1L": Classifier1L, "AL": ClassifierAL}


def topo_changes_experiments_helper(args, datasets):
    experiments = []
    num_of_hidden_layers = range(3, args.l + 3, 2)
    dim_of_hidden_layers = range(3, args.d + 3, 2)
    for n in num_of_hidden_layers:
        for d in dim_of_hidden_layers:
            experiments.append(
                TopologyChangeExperiments(
                    model=MODELS_DICT[args.m],
                    datasets=datasets,
                    model_config={
                        "num_of_hidden": n,
                        "dim_of_hidden": d,
                    },
                    n_experiments=args.ne,
                    list_of_activations=args.act,
                )
            )
    return experiments


def main():
    parser = argparse.ArgumentParser(
        description="Run the experiments for the Masters thesis."
    )
    parser.add_argument(
        "--e",
        type=str,
        help="which experiment to choose",
        default="topo_change",
        choices=["activation", "topo_change"],
    )
    parser.add_argument(
        "--l", type=int, default=5, help="num of layers to be used in a model"
    )
    parser.add_argument(
        "--d",
        type=int,
        default=8,
        help="dimension of hidden layers to be used in a model",
    )
    parser.add_argument(
        "--data",
        type=str,
        action="append",
        default=["circles", "tori", "disks"],
        choices=["circles", "tori", "disks"],
        help="datasets names to be used in the experiment",
    )
    parser.add_argument(
        "--act",
        type=str,
        action="append",
        default=["split_tanh", "relu"],
        choices=["split_tanh", "split_sign", "split_sincos", "relu"],
        help="activation functions names to be used in the experiment",
    )
    parser.add_argument(
        "--ne",
        type=int,
        default=30,
        help="number of experiments to play for each model",
    )
    parser.add_argument(
        "--m",
        type=str,
        default="AL",
        choices=["1L", "AL"],
        help="Which model to choose in the experiments",
    )

    args = parser.parse_args()

    datasets = [DATASETS_DICT[dataset]() for dataset in args.data]
    if args.e == "activation":
        experiment = EXPERIMENTS_DICT[args.e](
            model=MODELS_DICT[args.m],
            datasets=datasets,
            n_experiments=args.ne,
            num_of_hidden_layers=range(3, args.l + 3, 2),
            dim_of_hidden_layers=range(3, args.d + 3, 2),
            list_of_activations=args.act,
            verbose=False,
        )
        start_time = time.time()
        experiment.run_experiments()
        end_time = time.time()
    else:
        experiments = topo_changes_experiments_helper(args, datasets)
        start_time = time.time()
        for experiment in experiments:
            experiment.run_experiments()
        end_time = time.time()

    print("Time spent = {:.2f} min".format((end_time - start_time) / 60))


if __name__ == "__main__":
    main()