File size: 15,585 Bytes
e143b4a | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 | """Validate and ingest Karthik's EmpathRAG Core dataset delivery.
This script is intentionally conservative: it validates the delivery folder,
copies the single-turn labels into the supervised router dataset, copies the
multi-turn scenarios into the eval harness, and writes a small report. Resource
profile additions are validated and reported, but not auto-merged into the
runtime registry because contact details need a human provenance review.
"""
from __future__ import annotations
import argparse
import csv
import json
from pathlib import Path
import shutil
import sys
from typing import Any
ROOT = Path(__file__).resolve().parents[1]
sys.path.insert(0, str(ROOT / "src"))
from pipeline.v2_schema import SafetyTier, SupportRoute # noqa: E402
DEFAULT_DELIVERY = ROOT / "Data_Karthik" / "empathrag_core_dataset_v2"
DEFAULT_SINGLE_OUTPUT = ROOT / "eval" / "empathrag_core_supervised.csv"
DEFAULT_MULTITURN_OUTPUT = ROOT / "eval" / "multiturn_scenarios.jsonl"
DEFAULT_REPORT_JSON = ROOT / "eval" / "core_dataset_v2_ingest_report.json"
DEFAULT_REPORT_MD = ROOT / "eval" / "core_dataset_v2_ingest_report.md"
REQUIRED_FILES = {
"README_dataset_notes.md",
"single_turn_labeled.csv",
"multi_turn_scenarios.jsonl",
"source_target_map.csv",
"risky_ambiguous_cases.csv",
"resource_profile_additions.csv",
}
SINGLE_TURN_COLUMNS = [
"query_id",
"query_text",
"audience_mode",
"route_label",
"safety_tier",
"should_intercept",
"expected_usage_modes",
"preferred_source_names",
"avoid_source_names",
"preferred_topics",
"expected_response_action",
"tricky_flags",
"split",
"notes",
]
RESOURCE_ADDITION_COLUMNS = [
"resource_name",
"resource_type",
"official_url",
"source_authority",
"route_labels",
"safety_tiers",
"usage_modes",
"audience",
"contact_mode",
"contact_value",
"hours",
"location",
"confidentiality_status",
"last_verified",
"notes",
]
VALID_ROUTES = {item.value for item in SupportRoute}
VALID_TIERS = {item.value for item in SafetyTier}
VALID_AUDIENCE = {"student", "helping_friend"}
VALID_SPLITS = {"train", "dev", "test"}
VALID_USAGE_MODES = {"retrieval", "wellbeing_only", "crisis_only"}
def read_csv(path: Path) -> list[dict[str, str]]:
with path.open("r", encoding="utf-8-sig", newline="") as handle:
return list(csv.DictReader(handle))
def write_csv(path: Path, rows: list[dict[str, str]], fieldnames: list[str]) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
with path.open("w", encoding="utf-8", newline="") as handle:
writer = csv.DictWriter(handle, fieldnames=fieldnames, extrasaction="ignore")
writer.writeheader()
writer.writerows(rows)
def normalize_bool(value: str) -> str:
normalized = str(value).strip().lower()
if normalized in {"true", "yes", "1", "y"}:
return "true"
if normalized in {"false", "no", "0", "n"}:
return "false"
return normalized
def split_tokens(value: str) -> list[str]:
return [token.strip() for token in str(value).replace(";", ",").split(",") if token.strip()]
def validate_delivery(delivery_dir: Path) -> dict[str, Any]:
report: dict[str, Any] = {
"delivery_dir": str(delivery_dir),
"status": "pass",
"errors": [],
"warnings": [],
"counts": {},
"label_distribution": {"route_label": {}, "safety_tier": {}, "split": {}},
"next_steps": [],
}
if not delivery_dir.exists():
report["errors"].append(f"Delivery directory not found: {delivery_dir}")
report["status"] = "fail"
return report
missing = sorted(name for name in REQUIRED_FILES if not (delivery_dir / name).exists())
if missing:
report["errors"].append(f"Missing required files: {', '.join(missing)}")
single_path = delivery_dir / "single_turn_labeled.csv"
if single_path.exists():
rows = read_csv(single_path)
report["counts"]["single_turn_rows"] = len(rows)
_validate_single_turn_rows(rows, report)
multiturn_path = delivery_dir / "multi_turn_scenarios.jsonl"
if multiturn_path.exists():
scenarios = _read_jsonl(multiturn_path, report)
report["counts"]["multi_turn_scenarios"] = len(scenarios)
_validate_multiturn_scenarios(scenarios, report)
risky_path = delivery_dir / "risky_ambiguous_cases.csv"
if risky_path.exists():
report["counts"]["risky_ambiguous_rows"] = len(read_csv(risky_path))
resource_path = delivery_dir / "resource_profile_additions.csv"
if resource_path.exists():
additions = read_csv(resource_path)
report["counts"]["resource_profile_additions"] = len(additions)
_validate_resource_additions(additions, report)
if report["errors"]:
report["status"] = "fail"
elif report["warnings"]:
report["status"] = "pass_with_warnings"
report["next_steps"] = _next_steps(report)
return report
def _validate_single_turn_rows(rows: list[dict[str, str]], report: dict[str, Any]) -> None:
if not rows:
report["errors"].append("single_turn_labeled.csv has no rows.")
return
columns = set(rows[0].keys())
missing_columns = [col for col in SINGLE_TURN_COLUMNS if col not in columns]
if missing_columns:
report["errors"].append(f"single_turn_labeled.csv missing columns: {', '.join(missing_columns)}")
seen_ids: set[str] = set()
for idx, row in enumerate(rows, start=2):
query_id = row.get("query_id", "").strip()
if not query_id:
report["errors"].append(f"Row {idx}: query_id is blank.")
elif query_id in seen_ids:
report["errors"].append(f"Row {idx}: duplicate query_id {query_id}.")
seen_ids.add(query_id)
if not row.get("query_text", "").strip():
report["errors"].append(f"Row {idx}: query_text is blank.")
_validate_choice(row, "audience_mode", VALID_AUDIENCE, idx, report)
_validate_choice(row, "route_label", VALID_ROUTES, idx, report)
_validate_choice(row, "safety_tier", VALID_TIERS, idx, report)
_validate_choice(row, "split", VALID_SPLITS, idx, report)
should_intercept = normalize_bool(row.get("should_intercept", ""))
if should_intercept not in {"true", "false"}:
report["errors"].append(f"Row {idx}: should_intercept must be true/false/yes/no.")
row["should_intercept"] = should_intercept
for usage_mode in split_tokens(row.get("expected_usage_modes", "")):
if usage_mode not in VALID_USAGE_MODES:
report["warnings"].append(f"Row {idx}: unknown expected_usage_modes value '{usage_mode}'.")
if row.get("safety_tier") == SafetyTier.IMMINENT_SAFETY.value and should_intercept != "true":
report["warnings"].append(f"Row {idx}: imminent_safety row is not marked should_intercept=true.")
if should_intercept == "true" and row.get("safety_tier") != SafetyTier.IMMINENT_SAFETY.value:
report["warnings"].append(f"Row {idx}: intercept row should usually use imminent_safety tier.")
for field in ("route_label", "safety_tier", "split"):
value = row.get(field, "").strip()
bucket = report["label_distribution"][field]
bucket[value] = bucket.get(value, 0) + 1
def _validate_multiturn_scenarios(scenarios: list[dict[str, Any]], report: dict[str, Any]) -> None:
case_ids: set[str] = set()
for idx, scenario in enumerate(scenarios, start=1):
case_id = str(scenario.get("case_id", "")).strip()
if not case_id:
report["errors"].append(f"Scenario line {idx}: case_id is blank.")
elif case_id in case_ids:
report["errors"].append(f"Scenario line {idx}: duplicate case_id {case_id}.")
case_ids.add(case_id)
audience_mode = str(scenario.get("audience_mode", "student")).strip()
if audience_mode not in VALID_AUDIENCE:
report["errors"].append(f"Scenario {case_id or idx}: invalid audience_mode '{audience_mode}'.")
turns = scenario.get("turns")
if not isinstance(turns, list) or not turns:
report["errors"].append(f"Scenario {case_id or idx}: turns must be a non-empty list.")
continue
if len(turns) < 3:
report["warnings"].append(f"Scenario {case_id or idx}: fewer than 3 turns limits trajectory evaluation.")
for turn_idx, turn in enumerate(turns, start=1):
prefix = f"Scenario {case_id or idx} turn {turn_idx}"
if not str(turn.get("user", "")).strip():
report["errors"].append(f"{prefix}: user text is blank.")
route = str(turn.get("expected_route", "")).strip()
tier = str(turn.get("expected_safety_tier", "")).strip()
if route not in VALID_ROUTES:
report["errors"].append(f"{prefix}: invalid expected_route '{route}'.")
if tier not in VALID_TIERS:
report["errors"].append(f"{prefix}: invalid expected_safety_tier '{tier}'.")
intercept = normalize_bool(str(turn.get("should_intercept", "")))
if intercept not in {"true", "false"}:
report["errors"].append(f"{prefix}: should_intercept must be true/false.")
def _validate_resource_additions(rows: list[dict[str, str]], report: dict[str, Any]) -> None:
if not rows:
report["warnings"].append("resource_profile_additions.csv is empty.")
return
columns = set(rows[0].keys())
missing_columns = [col for col in RESOURCE_ADDITION_COLUMNS if col not in columns]
if missing_columns:
report["errors"].append(f"resource_profile_additions.csv missing columns: {', '.join(missing_columns)}")
for idx, row in enumerate(rows, start=2):
if not row.get("resource_name", "").strip():
report["errors"].append(f"Resource row {idx}: resource_name is blank.")
official_url = row.get("official_url", "").strip()
if not official_url.startswith(("https://", "http://", "internal://")):
report["errors"].append(f"Resource row {idx}: official_url must be a URL or internal:// provenance.")
for route in split_tokens(row.get("route_labels", "")):
if route not in VALID_ROUTES:
report["warnings"].append(f"Resource row {idx}: unknown route_labels value '{route}'.")
for tier in split_tokens(row.get("safety_tiers", "")):
if tier not in VALID_TIERS:
report["warnings"].append(f"Resource row {idx}: unknown safety_tiers value '{tier}'.")
for field in ("contact_value", "hours", "location"):
if not row.get(field, "").strip():
report["warnings"].append(f"Resource row {idx}: {field} is blank; use 'unknown' if not verified.")
def _validate_choice(
row: dict[str, str],
field: str,
valid_values: set[str],
row_number: int,
report: dict[str, Any],
) -> None:
value = row.get(field, "").strip()
if value not in valid_values:
report["errors"].append(f"Row {row_number}: invalid {field} '{value}'.")
def _read_jsonl(path: Path, report: dict[str, Any]) -> list[dict[str, Any]]:
rows: list[dict[str, Any]] = []
with path.open("r", encoding="utf-8") as handle:
for line_no, line in enumerate(handle, start=1):
if not line.strip():
continue
try:
value = json.loads(line)
except json.JSONDecodeError as exc:
report["errors"].append(f"{path.name} line {line_no}: invalid JSON: {exc}")
continue
if not isinstance(value, dict):
report["errors"].append(f"{path.name} line {line_no}: expected JSON object.")
continue
rows.append(value)
return rows
def _next_steps(report: dict[str, Any]) -> list[str]:
if report["status"] == "fail":
return [
"Send the report errors back to Karthik before training.",
"Do not overwrite eval/empathrag_core_supervised.csv with a failed delivery.",
]
return [
"Run eval/train_ml_router.py to train local TF-IDF route and tier classifiers.",
"Run eval/run_router_eval.py for Eval A single-turn router metrics.",
"Run eval/run_multiturn_eval.py for Eval B trajectory metrics.",
"Manually review resource_profile_additions.csv before merging new services into data/curated/service_graph.jsonl.",
]
def write_report(report: dict[str, Any], json_path: Path, md_path: Path) -> None:
json_path.parent.mkdir(parents=True, exist_ok=True)
json_path.write_text(json.dumps(report, indent=2), encoding="utf-8")
lines = [
"# EmpathRAG Core Dataset V2 Ingest Report",
"",
f"- Status: `{report['status']}`",
f"- Delivery directory: `{report['delivery_dir']}`",
"",
"## Counts",
]
for key, value in report.get("counts", {}).items():
lines.append(f"- {key}: {value}")
lines.extend(["", "## Label Distribution"])
for field, buckets in report.get("label_distribution", {}).items():
lines.append(f"### {field}")
if buckets:
for key, value in sorted(buckets.items()):
lines.append(f"- `{key}`: {value}")
else:
lines.append("- No rows counted.")
lines.extend(["", "## Errors"])
lines.extend([f"- {item}" for item in report["errors"]] or ["- None"])
lines.extend(["", "## Warnings"])
lines.extend([f"- {item}" for item in report["warnings"]] or ["- None"])
lines.extend(["", "## Next Steps"])
lines.extend([f"- {item}" for item in report["next_steps"]])
md_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
def ingest(args: argparse.Namespace) -> dict[str, Any]:
report = validate_delivery(args.delivery_dir)
write_report(report, args.report_json, args.report_md)
if report["status"] == "fail":
raise SystemExit(f"Dataset delivery failed validation. See {args.report_md}")
single_rows = read_csv(args.delivery_dir / "single_turn_labeled.csv")
for row in single_rows:
row["should_intercept"] = normalize_bool(row.get("should_intercept", ""))
write_csv(args.output, single_rows, SINGLE_TURN_COLUMNS)
if not args.skip_multiturn_copy:
args.multiturn_output.parent.mkdir(parents=True, exist_ok=True)
shutil.copyfile(args.delivery_dir / "multi_turn_scenarios.jsonl", args.multiturn_output)
return report
def main() -> None:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--delivery-dir", type=Path, default=DEFAULT_DELIVERY)
parser.add_argument("--output", type=Path, default=DEFAULT_SINGLE_OUTPUT)
parser.add_argument("--multiturn-output", type=Path, default=DEFAULT_MULTITURN_OUTPUT)
parser.add_argument("--report-json", type=Path, default=DEFAULT_REPORT_JSON)
parser.add_argument("--report-md", type=Path, default=DEFAULT_REPORT_MD)
parser.add_argument("--skip-multiturn-copy", action="store_true")
args = parser.parse_args()
report = ingest(args)
print(json.dumps({"status": report["status"], "counts": report["counts"]}, indent=2))
print(f"Wrote supervised labels to {args.output}")
if not args.skip_multiturn_copy:
print(f"Wrote multi-turn scenarios to {args.multiturn_output}")
print(f"Wrote ingest report to {args.report_md}")
if __name__ == "__main__":
main()
|