File size: 4,212 Bytes
3f316da | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 | #!/usr/bin/env python3
"""Run the prespecified five-stock hyperparameter sensitivity grid."""
from __future__ import annotations
import argparse
import copy
import json
import sys
from pathlib import Path
ROOT = Path(__file__).resolve().parents[3]
SCRIPT_DIR = ROOT / "experiments/tbd_revision/scripts"
if str(SCRIPT_DIR) not in sys.path:
sys.path.insert(0, str(SCRIPT_DIR))
from train_representation_baselines import run_one, summarize # noqa: E402
STOCKS = ("asset_001", "asset_002", "asset_017", "asset_005", "asset_033")
METHODS = ("timae", "lobgrid_timae", "simlob2026")
SEEDS = (2026, 2027, 2028)
# Each axis changes one value from the retained configuration. The baseline
# additionally supplies the three label-budget conditions used elsewhere.
VARIANTS: dict[str, dict[str, float | int]] = {
"baseline": {},
"latent64": {"latent_dim": 64},
"latent256": {"latent_dim": 256},
"width64": {"d_model": 64},
"width128": {"d_model": 128},
"heads2": {"nhead": 2},
"heads8": {"nhead": 8},
"mask20": {"mask_ratio": 0.2},
"mask60": {"mask_ratio": 0.6},
"lr100": {"learning_rate": 0.0001},
"lr1000": {"learning_rate": 0.001},
"batch64": {"batch_size": 64},
"batch256": {"batch_size": 256},
"wd0": {"weight_decay": 0.0},
"wd1e4": {"weight_decay": 0.0001},
}
def parse_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--base-config", type=Path, required=True)
parser.add_argument("--output-root", type=Path, required=True)
parser.add_argument(
"--variants",
nargs="+",
choices=tuple(VARIANTS),
required=True,
)
parser.add_argument("--force", action="store_true")
return parser.parse_args()
def resolved(path: Path) -> Path:
return path if path.is_absolute() else ROOT / path
def main() -> None:
args = parse_args()
base = json.loads(resolved(args.base_config).read_text(encoding="utf-8"))
output_root = resolved(args.output_root)
output_root.mkdir(parents=True, exist_ok=True)
for variant in args.variants:
config = copy.deepcopy(base)
config["training"].setdefault("nhead", 4)
config["training"].setdefault("mask_ratio", 0.4)
config["training"].update(VARIANTS[variant])
result_dir = output_root / variant
config["results_dir"] = str(result_dir)
config["pretrain_id"] = f"final_revision_hp_{variant}"
config["stocks"] = list(STOCKS)
config["methods"] = list(METHODS)
config["settings"] = ["from_scratch", "fine_tune"]
config["seeds"] = list(SEEDS)
config["label_budgets"] = (
[0.25, 0.5, 1.0] if variant == "baseline" else [0.5]
)
config["purpose"] = (
"Prespecified five-stock, three-seed, one-factor-at-a-time "
f"sensitivity condition: {variant}."
)
result_dir.mkdir(parents=True, exist_ok=True)
(result_dir / "resolved_config.json").write_text(
json.dumps(config, indent=2) + "\n",
encoding="utf-8",
)
total = (
len(STOCKS)
* len(METHODS)
* len(SEEDS)
* len(config["label_budgets"])
)
completed = 0
for stock in STOCKS:
for method in METHODS:
for seed in SEEDS:
for label_budget in config["label_budgets"]:
completed += 1
print(
f"VARIANT {variant} {completed}/{total} "
f"{stock} {method} seed={seed} "
f"budget={label_budget}",
flush=True,
)
run_one(
config,
stock,
method,
seed,
label_budget,
config["settings"],
force=args.force,
)
summarize(config, f"hyperparameter_{variant}")
if __name__ == "__main__":
main()
|