scripts/run_experiments.py

from __future__ import annotations

import argparse
import json

from common import load_config
from dynafall.train_eval import evaluate_checkpoint, train_model


def main() -> None:
    ap = argparse.ArgumentParser()
    ap.add_argument("--dataset", required=True)
    ap.add_argument("--methods", nargs="+", default=["lstm", "stgcn", "agcn", "ctrgcn", "posec3d", "tcnte", "dynafall"])
    ap.add_argument("--epochs", type=int, default=None)
    ap.add_argument("--config", default="configs/default.yaml")
    ap.add_argument("--processed-dataset", default=None)
    ap.add_argument("--out-root", default=None)
    args = ap.parse_args()
    cfg = load_config(args.config)
    all_results = []
    for method in args.methods:
        out_dir = None
        if args.out_root:
            out_dir = f"{args.out_root}/{args.dataset}/{method}"
        print(f"==> Training {method}")
        train_model(args.dataset, method, cfg, epochs=args.epochs, out_dir=out_dir, processed_dataset=args.processed_dataset)
        result = evaluate_checkpoint(args.dataset, method, cfg, out_dir=out_dir, processed_dataset=args.processed_dataset)
        all_results.append(result)
        print(json.dumps(result, indent=2))
    print(json.dumps(all_results, indent=2))


if __name__ == "__main__":
    main()