File size: 2,258 Bytes
2e511b5 | 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 | """Inference JSONL I/O: faithful round-trip and CSV-scoring equivalence."""
import csv
import json
from legex.evaluation import normalise
from legex.utils import inference_record, read_inference_jsonl, write_inference_jsonl
COLUMNS = ["case_id", "link", "legal_subject_judgement", "dispute_value_nominal",
"trial_end_date", "plaintiffs_all_count", "model", "error"]
ROWS = [
{"case_id": "C-1", "link": "", "legal_subject_judgement": "Civil_Law",
"dispute_value_nominal": "nonpecuniary", "trial_end_date": "2023-05-04",
"plaintiffs_all_count": "2", "model": "harvey", "error": ""},
# refusal prose + unicode + empty cells stay verbatim / become null
{"case_id": "C-2", "link": "", "legal_subject_judgement": "",
"dispute_value_nominal": "None (the amount is not stated) N/A",
"trial_end_date": "", "plaintiffs_all_count": "", "model": "harvey",
"error": "श्रेणी"},
]
def test_empty_becomes_null_nonempty_stays_string(tmp_path):
rec = inference_record(ROWS[1], COLUMNS)
assert rec["link"] is None and rec["trial_end_date"] is None # empty -> null
assert rec["dispute_value_nominal"] == "None (the amount is not stated) N/A" # verbatim
assert rec["error"] == "श्रेणी"
def test_roundtripnormalises_identically_to_csv(tmp_path):
"""Reading JSONL and reading the equivalent CSV must yield identical scored cells."""
jsonl = tmp_path / "inf.jsonl"
write_inference_jsonl(jsonl, ROWS, COLUMNS)
csv_path = tmp_path / "inf.csv"
with csv_path.open("w", encoding="utf-8", newline="") as f:
w = csv.DictWriter(f, fieldnames=COLUMNS)
w.writeheader()
w.writerows(ROWS)
def scored(rows):
return [{k: normalise(v) for k, v in r.items()} for r in rows]
with csv_path.open(encoding="utf-8", newline="") as f:
csv_rows = list(csv.DictReader(f))
assert scored(read_inference_jsonl(jsonl)) == scored(csv_rows)
def test_jsonl_is_one_object_per_line(tmp_path):
p = tmp_path / "inf.jsonl"
write_inference_jsonl(p, ROWS, COLUMNS)
lines = p.read_text(encoding="utf-8").splitlines()
assert len(lines) == len(ROWS)
assert all(json.loads(line) for line in lines) # each line is valid JSON
|