File size: 3,562 Bytes
3f97185 | 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 | """Commands do not run evaluations unless `evaluate` is explicitly invoked."""
from __future__ import annotations
import argparse
import json
from pathlib import Path
def main():
parser = argparse.ArgumentParser(prog="clc-mini-bench", description="CLC NKp46 benchmark preparation and frozen-feature evaluation")
sub = parser.add_subparsers(dest="command", required=True)
build = sub.add_parser("build", help="Prepare annotations, deduplicated datasets, and family folds")
build.add_argument("--raw", type=Path, default=Path("data/raw/measurements.csv"))
build.add_argument("--output", type=Path, default=Path("data/processed"))
build.add_argument("--protocol", type=Path)
check = sub.add_parser("validate", help="Validate both prepared datasets and file checksums")
check.add_argument("--data-dir", type=Path, default=Path("data/processed"))
embeddings = sub.add_parser("embed", help="Extract frozen embeddings with ESM-2 or a custom adapter")
embeddings.add_argument("--dataset", type=Path, required=True)
embeddings.add_argument("--output", type=Path, required=True)
embeddings.add_argument("--adapter", help="Python module:function receiving label-free sequence records")
embeddings.add_argument("--model", default="facebook/esm2_t33_650M_UR50D")
embeddings.add_argument("--revision", help="Model repository revision or commit")
embeddings.add_argument("--batch-size", type=int, default=16)
embeddings.add_argument("--device", default="cpu")
evaluate = sub.add_parser("evaluate", help="Run grouped outer CV with nested Ridge model selection")
evaluate.add_argument("--dataset", type=Path, required=True)
feature_source = evaluate.add_mutually_exclusive_group(required=True)
feature_source.add_argument("--features", type=Path)
feature_source.add_argument("--composition-baseline", action="store_true")
evaluate.add_argument("--output", type=Path, required=True)
evaluate.add_argument("--protocol", type=Path)
args = parser.parse_args()
if args.command == "build":
from .data import build as run
result = run(args.raw, args.output, args.protocol)
print(json.dumps(result["datasets"], indent=2))
elif args.command == "validate":
import pandas as pd
from .data import sha256_file, validate_dataset
manifest = json.loads((args.data_dir / "manifest.json").read_text())
for name, expected in manifest["files"].items():
if sha256_file(args.data_dir / name) != expected:
raise ValueError(f"File checksum mismatch: {name}")
for platform in ("omniflic", "omniclic"):
frame = pd.read_csv(args.data_dir / f"{platform}_nkp46.csv")
validate_dataset(frame)
print(f"{platform}: {len(frame)} samples, {frame.family_id.nunique()} families, {frame.fold_id.nunique()} folds; valid")
elif args.command == "embed":
from .embeddings import extract_embeddings
result = extract_embeddings(args.dataset, args.output, adapter=args.adapter,
model_name=args.model, batch_size=args.batch_size, device=args.device, revision=args.revision)
print(json.dumps(result, indent=2))
elif args.command == "evaluate":
from .evaluation import evaluate as run
result = run(args.dataset, args.output, features=args.features,
baseline=args.composition_baseline, protocol_path=args.protocol)
print(json.dumps(result["pooled_out_of_fold"], indent=2))
if __name__ == "__main__":
main()
|