File size: 1,259 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 | """ecfr_dolma.py ECFR-titleN.xml OUT.json.gz (10-27) - govinfo eCFR bulk XML -> dolma json.gz for dolma_chunk.py doc: one doc per
SECTION (DIV8), title "N CFR § x.y Heading" (research/finance.md: questions name the citation, so BM25 needs it in the passage),
text = its paragraphs. Public domain."""
import sys, re, json, gzip
import xml.etree.ElementTree as ET
src, dst = sys.argv[1:3]
tn = re.search(r"title-?(\d+)", src).group(1)
txt = lambda e: re.sub(r"\s+", " ", "".join(e.itertext())).strip()
n, part = 0, ""
with gzip.open(dst + ".tmp", "wt") as f:
for ev, e in ET.iterparse(src, events=("end",)):
if e.tag.startswith("DIV") and e.get("TYPE") == "PART":
e.clear(); continue
if e.tag.startswith("DIV") and e.get("TYPE") == "SECTION":
h = e.find("HEAD"); head = txt(h) if h is not None else e.get("N", "")
body = "\n".join(txt(p) for p in e.iter() if p.tag in ("P", "FP", "NOTE", "EXTRACT") and txt(p))
if len(body.split()) >= 20:
f.write(json.dumps({"id": f"ecfr{tn}-{n}", "text": body, "metadata": {"title": f"{tn} CFR {head}"[:160]}}) + "\n"); n += 1
e.clear()
import os # noqa: E402
os.replace(dst + ".tmp", dst)
print(src, "sections", n, flush=True)
|