File size: 2,426 Bytes
db32e07 | 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 | #!/usr/bin/env python
"""
Build FAISS indices from library vectors. Plan §1.3: index_smi, index_chem.
"""
from __future__ import annotations
import argparse
import sys
from pathlib import Path
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
from spec_rag.faiss_index import build_hnsw_index, save_index
from spec_rag.io import ensure_dir
def parse_args():
p = argparse.ArgumentParser(description="Build index_smi.faiss, index_chem.faiss, and index_flare.faiss")
p.add_argument("--library-dir", required=True, help="Directory with vectors_smi.npy, vectors_chem.npy, and/or vectors_flare.npy")
p.add_argument("--out-dir", default=None, help="Defaults to library-dir")
p.add_argument("--metric", choices=["cosine", "l2"], default="cosine")
p.add_argument("--m", type=int, default=32)
p.add_argument("--ef-construction", type=int, default=200)
p.add_argument("--ef-search", type=int, default=256, help="HNSW ef_search; higher improves recall")
p.add_argument("--smi-only", action="store_true", help="Only build index_smi")
p.add_argument("--chem-only", action="store_true", help="Only build index_chem")
p.add_argument("--flare-only", action="store_true", help="Only build index_flare")
return p.parse_args()
def main():
args = parse_args()
lib = Path(args.library_dir)
out = ensure_dir(args.out_dir or lib)
import numpy as np
if args.flare_only:
targets = [("flare", "vectors_flare.npy", "index_flare.faiss")]
else:
targets = []
if not args.chem_only:
targets.append(("smi", "vectors_smi.npy", "index_smi.faiss"))
targets.append(("flare", "vectors_flare.npy", "index_flare.faiss"))
if not args.smi_only:
targets.append(("chem", "vectors_chem.npy", "index_chem.faiss"))
for label, vec_name, index_name in targets:
vec_path = lib / vec_name
if not vec_path.exists():
print(f"Skip index_{label}: {vec_path} not found")
continue
v = np.load(vec_path).astype(np.float32)
idx = build_hnsw_index(
v, m=args.m, ef_construction=args.ef_construction,
ef_search=args.ef_search, metric=args.metric,
)
save_index(idx, out / index_name)
print(f"Saved {index_name} ({v.shape[0]} vectors)")
if __name__ == "__main__":
main()
|