File size: 2,183 Bytes
901f964 | 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 | """bm25.py TABLE - lexical (BM25) first stage for an engram table (10-20). The static keys average token vectors, so rare exact
identifiers (json_encode flags, zoneinfo.TZPATH, case names, gene symbols) blur away; BM25 keeps them. tantivy index of the
passage text (en_stem, term frequencies only, no positions, nothing stored but the pid) in <table>/bm25/, built from the
table's passages/*.parquet. Resumable per passage file (bm25/done.txt). engram.py merges its hits with the static ones before
the re-rank. Env: THREADS (2), HEAP_MB (1000)."""
import os, sys, glob, time
import pyarrow.parquet as pq, tantivy
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
E = os.environ.get("ENGRAM_ROOT", ".")
def schema():
sb = tantivy.SchemaBuilder()
sb.add_text_field("text", stored=False, tokenizer_name="en_stem", index_option="freq")
sb.add_integer_field("pid", stored=True, indexed=False, fast=False)
return sb.build()
def open_index(root, create=False):
d = f"{root}/bm25"
if create:
os.makedirs(d, exist_ok=True)
return tantivy.Index(schema(), path=d, reuse=True)
if __name__ == "__main__":
t = sys.argv[1]; root = E if t == "wiki" else f"{E}/tables/{t}"
ix = open_index(root, create=True); donef = f"{root}/bm25/done.txt"
done = set(open(donef).read().split()) if os.path.exists(donef) else set()
w = ix.writer(heap_size=int(os.environ.get("HEAP_MB", 1000)) * 1_000_000, num_threads=int(os.environ.get("THREADS", 2)))
for f in sorted(glob.glob(f"{root}/passages/*.parquet")):
name = os.path.basename(f)
if name in done:
continue
t0 = time.time(); n = 0
for b in pq.ParquetFile(f).iter_batches(batch_size=8192, columns=["pid", "text"]):
for pid, text in zip(b.column("pid").to_pylist(), b.column("text").to_pylist()):
w.add_document(tantivy.Document(text=text, pid=pid)); n += 1
w.commit(); open(donef, "a").write(name + "\n")
print(f"{time.strftime('%H:%M:%S')} {t} {name}: {n} in {time.time() - t0:.0f} s", flush=True)
w.wait_merging_threads()
print(f"{time.strftime('%H:%M:%S')} {t} bm25 done", flush=True)
|