Datasets:
File size: 4,096 Bytes
b3991a8 | 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 | """Harvest the dev split of canary-eval from the Canary test suite.
Only checker calls whose verdict is explicitly asserted by a test are kept
(`self.assertEqual(check(...)["result"], expected)`), and the label is that asserted
expectation - never the engine's own output. Run from the repository root:
python hf/dataset/scripts/harvest_dev.py hf/dataset/data/dev.jsonl
"""
import copy
import json
import re
import sys
import unittest
sys.path.insert(0, ".")
import logging # noqa: E402
def main() -> None:
"""Runs the test suite with recording hooks and writes the asserted cases (the regex worker re-imports this module, so nothing may run at import time)."""
logging.disable(logging.CRITICAL)
import test_app
JA = re.compile(r"[-ヿ㐀-鿿]")
records: list[dict] = []
current_test = {"name": ""}
class Tagged(str):
"""A verdict string that remembers which recorded call produced it."""
class Rec(dict):
def __getitem__(self, key):
value = dict.__getitem__(self, key)
if key == "result" and isinstance(value, str):
tagged = Tagged(value)
tagged.rec = self.rec # type: ignore[attr-defined]
return tagged
return value
original_check = test_app.check
def recording_check(c, output, case=None, samples=None):
res = original_check(c, output, case, samples)
rec = {"test": current_test["name"], "contract": copy.deepcopy(c), "output": copy.deepcopy(output),
"case": copy.deepcopy(case or {}), "samples": copy.deepcopy(samples), "label": None}
records.append(rec)
out = Rec(res)
out.rec = rec
return out
test_app.check = recording_check
original_assert_equal = unittest.TestCase.assertEqual
def recording_assert_equal(self, first, second, msg=None):
if isinstance(first, Tagged) and isinstance(second, str) and second in ("pass", "fail", "skipped", "error"):
first.rec["label"] = second
return original_assert_equal(self, first, second, msg)
unittest.TestCase.assertEqual = recording_assert_equal
class Tracker(unittest.TextTestResult):
def startTest(self, test):
current_test["name"] = test.id().split(".", 1)[1]
super().startTest(test)
suite = unittest.defaultTestLoader.loadTestsFromModule(test_app)
result = unittest.TextTestRunner(resultclass=Tracker, verbosity=0, stream=open("/dev/null", "w")).run(suite)
assert result.wasSuccessful(), "the test suite must pass before harvesting"
SKIP_TESTS = ("TestPerformance", "test_whitespace_heavy", "TestNeverRaises", "regex_guard", "regex_worker", "timeouts_do_not")
seen = set()
rows = []
for r in records:
if r["label"] not in ("pass", "fail", "skipped") or any(s in r["test"] for s in SKIP_TESTS):
continue
contract = r["contract"]
output = r["samples"] if r["samples"] else r["output"]
key = json.dumps([contract, output, r["case"]], sort_keys=True, ensure_ascii=False)
if key in seen:
continue
seen.add(key)
text = json.dumps([output, r["case"]], ensure_ascii=False)
rows.append({
"id": f"dev-{len(rows) + 1:04d}",
"contract_type": contract["type"],
"contract": {k: v for k, v in contract.items() if k not in ("id", "severity")},
"output": output,
"case": r["case"],
"label": r["label"],
"lang": "ja" if JA.search(text) else "en",
"source": "test_app.py::" + r["test"],
})
with open(sys.argv[1], "w", encoding="utf-8") as fh:
for row in rows:
fh.write(json.dumps(row, ensure_ascii=False) + "\n")
by_type = {}
for row in rows:
by_type.setdefault(row["contract_type"], {"pass": 0, "fail": 0, "skipped": 0})[row["label"]] += 1
print(len(rows), "dev cases")
for t, c in sorted(by_type.items()):
print(f" {t:18s} {c}")
if __name__ == "__main__":
main()
|