Datasets:
File size: 3,532 Bytes
29d94b3 | 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 | #!/usr/bin/env python3
"""Fast open-set eval after gold/paradigm/booster upgrades."""
from __future__ import annotations
import json
from collections import Counter, defaultdict
from datetime import datetime, timezone
from pathlib import Path
from promote_and_train_classical import inject, load_all_gold, split_rows
from held_out_classical import score
from PFLT_FSOT_2_1_aligned import PFLT
OUT = Path(__file__).resolve().parent / "data" / "fast_open_eval_report.json"
def main() -> None:
gold = load_all_gold()
train, test = split_rows(gold, 0.85)
print(f"gold={len(gold)} train={len(train)} test={len(test)}", flush=True)
p = PFLT(
load_historical=True,
load_classical=False,
load_hieroglyphs=False,
load_domain_lexica=False,
enable_gapfill=True,
)
n = inject(p, train, expand_paradigms=True)
print(f"injected≈{n} pul_terms={len(p.pul_terms)}", flush=True)
# train closed sample
tr = score(p, train[:: max(1, len(train) // 1500)][:1500])
print(f"train_closed exact={tr['exact_rate']*100:.2f}% n={tr['n']}", flush=True)
# full test open
te = score(p, test)
print(
f"test_open exact={te['exact_rate']*100:.2f}% "
f"partial={te['exact_or_partial_rate']*100:.2f}% n={te['n']}",
flush=True,
)
by = defaultdict(list)
for d in te.get("misses", [])[:5]:
pass
# per-lang via re-score groups (only large langs)
from held_out_classical import score as sc
by_lang = defaultdict(list)
for r in test:
by_lang[r["source_lang"]].append(r)
lang_stats = {}
for lang, rows in sorted(by_lang.items(), key=lambda x: -len(x[1])):
if len(rows) < 30:
continue
s = sc(p, rows)
lang_stats[lang] = {
"n": s["n"],
"exact": s["exact_rate"],
"partial": s["exact_or_partial_rate"],
}
print(
f" {lang:5s} n={s['n']:4d} exact={s['exact_rate']*100:5.2f}% "
f"partial={s['exact_or_partial_rate']*100:5.2f}%",
flush=True,
)
# no-gap ablation on sample
p0 = PFLT(
load_historical=True,
load_classical=False,
load_hieroglyphs=False,
load_domain_lexica=False,
enable_gapfill=False,
)
inject(p0, train, expand_paradigms=False)
te0 = sc(p0, test[::3][:800])
print(
f"ablation no_gap+no_paradigm sample n={te0['n']} "
f"exact={te0['exact_rate']*100:.2f}% partial={te0['exact_or_partial_rate']*100:.2f}%",
flush=True,
)
report = {
"built_utc": datetime.now(timezone.utc).isoformat(),
"n_gold": len(gold),
"n_train": len(train),
"n_test": len(test),
"pul_terms": len(p.pul_terms),
"train_closed": {"exact": tr["exact_rate"], "partial": tr["exact_or_partial_rate"], "n": tr["n"]},
"test_open": {
"exact": te["exact_rate"],
"partial": te["exact_or_partial_rate"],
"n": te["n"],
"hits_sample": te.get("hits_sample", [])[:10],
"misses_sample": te.get("misses", [])[:10],
},
"by_lang": lang_stats,
"ablation_no_gap_sample": {
"exact": te0["exact_rate"],
"partial": te0["exact_or_partial_rate"],
"n": te0["n"],
},
}
OUT.write_text(json.dumps(report, indent=2, ensure_ascii=False), encoding="utf-8")
print("wrote", OUT, flush=True)
if __name__ == "__main__":
main()
|