File size: 5,443 Bytes
90264a9 | 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 | #!/usr/bin/env python3
"""Assemble the repo folder data_synthetic_extended/ (data part) from runs/ -> publish/.
python publish.py # variants/, generated/, wiki_check/, inventory/ (+ qa.py writes stats.json, QA_REPORT.md)
- variants/<epoch>/<type>_<k>.jsonl one file per data_synthetic chunk, same order as the originals, id <parent>-w1
- generated/<epoch>/<kind>_<k>.jsonl kind = mcq | tf | open | essay, 40 tasks per file (20 for essays),
'Zadanie N.' renumbered 1..40 inside each file
- wiki_check/<epoch>.jsonl fact-check of every original task against Wikipedia (last record per id)
- inventory/<epoch>.jsonl requirement -> Wikipedia articles (+ passages) and essay foci per chapter
Internal fields (_file, pid) are dropped.
"""
import re
from collections import defaultdict
from common import EPOCHS, PUBLISH, RUNS, load_originals, read_jsonl, write_jsonl
HDR = re.compile(r"^Zadanie \d+\.")
TF_I = re.compile(r"Oceń prawdziwość|Zaznacz P|Wybierz P,|jeśli zdanie jest prawdziwe|prawdziwe, a które fałszywe", re.I)
STD_TF = {"Oceń prawdziwość zdań. Zaznacz P, jeśli zdanie jest prawdziwe, albo F – jeśli jest fałszywe.",
"Na podstawie tekstu i własnej wiedzy oceń prawdziwość zdań. Zaznacz P, jeśli zdanie jest prawdziwe, albo F – "
"jeśli jest fałszywe."}
def fix_tf(r):
"""Variants written before the intro check: the intro already is an instruction -> drop the standard one."""
if r.get("task_type") != "true_false":
return r
lines = r["question"].split("\n")
if sum(1 for l in lines if TF_I.search(l)) >= 2:
for i in range(len(lines) - 1, 0, -1):
if lines[i] in STD_TF:
del lines[i]
if i < len(lines) and i > 0 and not lines[i - 1].strip() and not lines[i].strip():
del lines[i]
break
r = {**r, "question": "\n".join(lines)}
return r
def last_by(rows, key):
d = {}
for r in rows:
d[r[key]] = r
return d
def clean(r):
return {k: v for k, v in r.items() if not k.startswith("_")}
def main():
order = {r["id"]: (r["_file"], i) for i, r in enumerate(load_originals())}
# variants
var = last_by(read_jsonl(RUNS / "variants" / "variants.jsonl"), "parent_id")
files = defaultdict(list)
for pid, v in var.items():
f, i = order[pid]
files[f].append((i, fix_tf(clean(v))))
for d in (PUBLISH / "variants").glob("*/*.jsonl"):
d.unlink()
for f, rows in files.items():
ep, name = f.split("/")
write_jsonl(PUBLISH / "variants" / ep / name.replace(".verified", ""), [r for _, r in sorted(rows, key=lambda x: x[0])])
# wiki_check
wc = last_by(read_jsonl(RUNS / "variants" / "wiki_check.jsonl"), "id")
for ep in EPOCHS:
rows = [wc[k] for k in sorted(wc, key=lambda k: order.get(k, ("", 0))) if k.startswith(f"syn-{ep}-")]
if rows:
write_jsonl(PUBLISH / "wiki_check" / f"{ep}.jsonl", rows)
# generated
rows = read_jsonl(RUNS / "generate" / "generated.jsonl")
for r in rows: # early essay ids lacked the chapter (collisions): rebuild ids from seed.unit
kind = r["id"].split("-")[2]
r["id"] = f"wik-{r['epoch']}-{kind}-{'-'.join(r['seed']['unit'].split('-')[2:])}-{r['id'].rsplit('-', 1)[-1]}"
gen = last_by(rows, "id")
by = defaultdict(list)
for r in gen.values():
kind = r["id"].split("-")[2]
by[(r["epoch"], kind)].append(r)
for d in (PUBLISH / "generated").glob("*/*.jsonl"):
d.unlink()
n_gen = 0
for (ep, kind), rows in sorted(by.items()):
rows.sort(key=lambda r: r["id"])
size = 20 if kind == "essay" else 40
for k in range(0, len(rows), size):
chunk = []
for j, r in enumerate(rows[k:k + size]):
r = dict(r)
r["question"] = HDR.sub(f"Zadanie {j + 1}.", r["question"], count=1)
chunk.append(r)
write_jsonl(PUBLISH / "generated" / ep / f"{kind}_{k // size}.jsonl", chunk)
n_gen += len(chunk)
# inventory
reqs = read_jsonl(RUNS / "inventory" / "reqs.jsonl")
foci = read_jsonl(RUNS / "inventory" / "foci.jsonl")
for ep in EPOCHS:
rows = []
for r in (x for x in reqs if x["epoch"] == ep):
rows.append({"type": "requirement", **{k: r[k] for k in ("req", "chapter", "chapter_title", "text", "issues")},
"articles": [{k: a.get(k) for k in ("title", "url", "snap", "kind", "purpose", "passages")}
for a in r["articles"]]})
for f in (x for x in foci if x["epoch"] == ep):
rows.append({"type": "essay_focus", **{k: f.get(k) for k in ("focus_id", "chapter", "chapter_title", "focus",
"period", "requirements", "angles")},
"articles": [{k: a.get(k) for k in ("title", "url", "snap", "passages")} for a in f["articles"]]})
if rows:
write_jsonl(PUBLISH / "inventory" / f"{ep}.jsonl", rows)
print(f"published: variants {len(var)} in {len(files)} files, wiki_check {len(wc)}, generated {n_gen}, "
f"inventory {len(reqs)} requirements + {len(foci)} foci -> {PUBLISH}")
if __name__ == "__main__":
main()
|