File size: 7,456 Bytes
5017b5d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 | #!/usr/bin/env python3
"""
Independent validation pass over the generated ICSThreatQA Hugging Face
table. Re-derives the dataset from the original .xlsx files and compares it
against the checked-in CSV/Parquet outputs, then reports data-quality
findings without silently modifying anything.
Usage:
python scripts/validate_dataset.py
Exit code is non-zero only for an *actual* data error (missing file, missing
column, unexpected row-count drift, duplicate IDs, CSV/Parquet mismatch, or
non-deterministic regeneration). Duplicate questions/rows that genuinely
exist in the original benchmark, and the known trailing-blank-row formatting
artifact in factual.xlsx, are reported but do NOT fail validation.
"""
from __future__ import annotations
import sys
from pathlib import Path
import pandas as pd
SCRIPT_DIR = Path(__file__).resolve().parent
sys.path.insert(0, str(SCRIPT_DIR))
import prepare_dataset as prep # noqa: E402
RELEASE_DIR = SCRIPT_DIR.parent
DATA_DIR = RELEASE_DIR / "data"
CSV_PATH = DATA_DIR / "icsthreatqa.csv"
PARQUET_PATH = DATA_DIR / "icsthreatqa.parquet"
EXPECTED_TOTAL = 620
def section(title: str) -> None:
print()
print(f"--- {title} ---")
def main() -> int:
errors: list[str] = []
notes: list[str] = []
if not CSV_PATH.exists():
print(f"VALIDATION FAILED: {CSV_PATH} not found. Run prepare_dataset.py first.")
return 1
if not PARQUET_PATH.exists():
print(f"VALIDATION FAILED: {PARQUET_PATH} not found. Run prepare_dataset.py first.")
return 1
csv_df = pd.read_csv(CSV_PATH, dtype=str, keep_default_na=False)
parquet_df = pd.read_parquet(PARQUET_PATH).astype(str)
# ------------------------------------------------------------------
# 1. Deterministic regeneration: rebuild independently from raw xlsx
# and confirm it matches the checked-in CSV exactly.
# ------------------------------------------------------------------
section("Deterministic regeneration")
frames = []
per_category_counts = {}
try:
for category, (filename, prefix) in prep.CATEGORIES.items():
df = prep.load_category(category, filename, prefix)
per_category_counts[category] = len(df)
frames.append(df)
rebuilt = pd.concat(frames, ignore_index=True)
except prep.ValidationError as e:
errors.append(f"Could not regenerate dataset from raw files: {e}")
rebuilt = None
if rebuilt is not None:
rebuilt_str = rebuilt.astype(str)
if rebuilt_str.equals(csv_df.reset_index(drop=True)):
print("OK: regenerating from raw .xlsx files reproduces the checked-in CSV exactly.")
else:
errors.append(
"Regenerating from raw .xlsx files does NOT match the checked-in CSV. "
"The dataset may be stale relative to the source files."
)
# ------------------------------------------------------------------
# 2. Record counts
# ------------------------------------------------------------------
section("Record counts")
total = len(csv_df)
print(f"Total QA records: {total}")
for cat, n in per_category_counts.items():
print(f" {cat}: {n}")
if total != EXPECTED_TOTAL:
notes.append(
f"Total record count is {total}, not the {EXPECTED_TOTAL} pairs described "
f"in the project README. Verified against raw .xlsx row counts (see below); "
f"this is reported as a discrepancy, not corrected."
)
print(
f"NOTE: expected {EXPECTED_TOTAL} per README, found {total}. "
f"Not auto-corrected -- see report."
)
else:
print(f"Matches expected total of {EXPECTED_TOTAL}.")
# ------------------------------------------------------------------
# 3. Missing questions / answers
# ------------------------------------------------------------------
section("Missing values")
missing_q = (csv_df["question"].str.strip() == "").sum()
missing_a = (csv_df["answer"].str.strip() == "").sum()
print(f"Missing/empty questions: {missing_q}")
print(f"Missing/empty answers: {missing_a}")
if missing_q or missing_a:
errors.append(f"Found missing question/answer fields: {missing_q} / {missing_a}")
# ------------------------------------------------------------------
# 4. Duplicates -- reported, not removed
# ------------------------------------------------------------------
section("Duplicate analysis (informational -- originals are preserved)")
dup_questions_overall = csv_df["question"].duplicated().sum()
print(f"Duplicate questions (exact text, across entire dataset): {dup_questions_overall}")
for cat in per_category_counts:
sub = csv_df[csv_df["question_type"] == cat]
d = sub["question"].duplicated().sum()
if d:
print(f" within {cat}: {d} duplicate question(s) -- present in original {cat}.xlsx")
dup_full_rows = csv_df.duplicated(subset=["question", "answer", "question_type"]).sum()
print(f"Duplicate full rows (question+answer+category): {dup_full_rows}")
notes.append(
f"{dup_questions_overall} duplicate question(s) and {dup_full_rows} duplicate "
f"full row(s) exist in the original benchmark workbooks and were intentionally "
f"preserved (not deduplicated)."
)
dup_ids = csv_df["id"].duplicated().sum()
print(f"Duplicate stable IDs: {dup_ids}")
if dup_ids:
errors.append(f"Found {dup_ids} duplicate stable ID(s) -- this IS an error.")
# ------------------------------------------------------------------
# 5. Category values
# ------------------------------------------------------------------
section("Category values")
valid_categories = set(prep.CATEGORIES.keys())
found_categories = set(csv_df["question_type"].unique())
print(f"question_type values found: {sorted(found_categories)}")
if found_categories - valid_categories:
errors.append(f"Unexpected question_type value(s): {found_categories - valid_categories}")
if valid_categories - found_categories:
errors.append(f"Missing expected question_type value(s): {valid_categories - found_categories}")
# ------------------------------------------------------------------
# 6. CSV / Parquet equality
# ------------------------------------------------------------------
section("CSV / Parquet equality")
parquet_df_filled = parquet_df.fillna("")
if csv_df.reset_index(drop=True).equals(parquet_df_filled.reset_index(drop=True)):
print("OK: CSV and Parquet contain identical data.")
else:
errors.append("CSV and Parquet outputs do not contain identical data.")
# ------------------------------------------------------------------
# Final report
# ------------------------------------------------------------------
section("Summary")
if notes:
print("Notes (non-fatal, documented discrepancies / known duplicates):")
for n in notes:
print(f" - {n}")
if errors:
print("\nERRORS (actual data-quality problems):")
for e in errors:
print(f" - {e}")
print("\nVALIDATION FAILED.")
return 1
print("\nVALIDATION PASSED (with the documented notes above, if any).")
return 0
if __name__ == "__main__":
raise SystemExit(main())
|