Spaces:
Sleeping
Sleeping
File size: 9,916 Bytes
ebc7457 |
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 |
"""
Full benchmark for NL2SQL pipeline.
Metrics:
- EM (exact match)
- Structural Match (sqlglot AST)
- Execution Accuracy
- Safety consistency (pipeline vs AST)
- Latency (end-to-end) + per-stage trace (via pipeline if available)
Outputs:
JSONL (logs), JSON (summary), CSV (compact table)
Run example:
python benchmarks/evaluate_spider_pro.py --limit 10 --sleep 0.1 --db sqlite --adapter data/chinook.db
"""
from __future__ import annotations
import argparse
import csv
import json
import sqlite3
import time
from pathlib import Path
from typing import Any, Dict, List, Optional, cast
import sqlglot
from sqlglot.errors import ParseError
# Reuse existing factories from FastAPI router (no new DI needed)
from app.routers.nl2sql import ( # type: ignore
_pipeline as DEFAULT_PIPELINE,
_build_pipeline,
_select_adapter,
)
from nl2sql.safety import Safety
# -------------------- Helpers --------------------
def _int_ms(start: float) -> int:
return int((time.perf_counter() - start) * 1000)
def _parse_sql(sql: str) -> Optional[sqlglot.Expression]:
try:
return sqlglot.parse_one(sql, read="sqlite")
except ParseError:
return None
def _is_structural_match(sql1: str, sql2: str) -> bool:
a, b = _parse_sql(sql1), _parse_sql(sql2)
return (a == b) if (a is not None and b is not None) else False
def _exec_sql(conn: sqlite3.Connection, sql: str) -> List[tuple]:
try:
cur = conn.execute(sql)
return [tuple(r) for r in cur.fetchall()]
except Exception:
return []
def _derive_schema_preview_safe(pipeline_obj: Any) -> Optional[str]:
for attr in ("executor", "adapter"):
obj = getattr(pipeline_obj, attr, None)
if obj is not None and hasattr(obj, "derive_schema_preview"):
try:
# type: ignore[no-any-return]
return obj.derive_schema_preview() # pragma: no cover
except Exception:
pass
return None
def _to_stage_list(trace_obj: Any) -> List[Dict[str, Any]]:
"""
Normalize pipeline trace (list of dataclass or dict) to:
[{'stage': str, 'ms': int}, ...]
"""
stages: List[Dict[str, Any]] = []
if not isinstance(trace_obj, list):
return stages
for t in trace_obj:
if isinstance(t, dict):
stage = t.get("stage", "?")
ms = t.get("duration_ms", 0)
else:
stage = getattr(t, "stage", "?")
ms = getattr(t, "duration_ms", 0)
try:
stages.append({"stage": str(stage), "ms": int(ms)})
except Exception:
stages.append({"stage": str(stage), "ms": 0})
return stages
# -------------------- Main --------------------
def main() -> None:
parser = argparse.ArgumentParser()
parser.add_argument("--limit", type=int, default=10, help="Max number of examples")
parser.add_argument("--resume", type=int, default=0, help="Skip first N examples")
parser.add_argument(
"--sleep", type=float, default=0.0, help="Delay (seconds) between queries"
)
parser.add_argument(
"--split", type=str, default="test", help="Dataset split (placeholder)"
)
parser.add_argument(
"--db", type=str, default="sqlite", help="Database ID (e.g., sqlite/postgres)"
)
parser.add_argument(
"--adapter",
type=str,
default="data/chinook.db",
help="SQLite file path for local eval",
)
args = parser.parse_args()
# SQLite connection for execution-accuracy
conn = sqlite3.connect(args.adapter)
# Build pipeline from router factories
try:
adapter = _select_adapter(args.db)
pipeline = _build_pipeline(adapter)
using_default = False
except Exception:
pipeline = DEFAULT_PIPELINE
using_default = True
safety = Safety()
schema_preview = _derive_schema_preview_safe(pipeline)
print(f"β
Pipeline ready (db={args.db}, default={using_default})")
# Minimal sample dataset for demonstration; replace with real Spider subset if available
DATASET: List[Dict[str, Any]] = [
{
"id": 1,
"question": "list all customers",
"gold_sql": "SELECT * FROM customers;",
},
{
"id": 2,
"question": "top 3 albums by total sales",
"gold_sql": """
SELECT a.Title, SUM(i.Quantity * i.UnitPrice) AS total
FROM albums a
JOIN tracks t ON a.AlbumId = t.AlbumId
JOIN invoice_items i ON t.TrackId = i.TrackId
GROUP BY a.AlbumId
ORDER BY total DESC
LIMIT 3;
""",
},
{
"id": 3,
"question": "number of employees per city",
"gold_sql": """
SELECT City, COUNT(*) AS cnt
FROM employees
GROUP BY City
ORDER BY cnt DESC;
""",
},
]
sliced = DATASET[args.resume : args.resume + args.limit]
# Eval loop
results: List[Dict[str, Any]] = []
for idx, ex in enumerate(sliced, start=1):
qid = cast(int, ex.get("id", idx))
q: str = cast(str, ex.get("question", ""))
gold_sql: str = cast(str, ex.get("gold_sql", "")).strip()
print(f"\n[{idx}] {q}")
t0 = time.perf_counter()
try:
out = pipeline.run(user_query=q, schema_preview=(schema_preview or "")) # type: ignore[misc]
latency = _int_ms(t0)
# Safely extract predicted SQL:
sql_pred_obj = getattr(out, "sql", None)
if sql_pred_obj is None:
data_obj = getattr(out, "data", None)
if data_obj is not None:
sql_pred_obj = getattr(data_obj, "sql", None)
sql_pred: str = str(sql_pred_obj) if sql_pred_obj is not None else ""
if not sql_pred.strip():
raise ValueError("No SQL generated")
# Metrics
em = sql_pred.strip().lower() == gold_sql.strip().lower()
sm = _is_structural_match(sql_pred, gold_sql)
safe_ast = safety.check(sql_pred) # pipeline has its own safety as well
safe_pipeline = bool(getattr(out, "ok", True))
safety_consistent = safe_ast.ok == safe_pipeline
gold_exec = _exec_sql(conn, gold_sql)
pred_exec = _exec_sql(conn, sql_pred)
exec_acc = gold_exec == pred_exec
stages = _to_stage_list(getattr(out, "trace", None))
results.append(
{
"id": qid,
"question": q,
"sql_pred": sql_pred,
"sql_gold": gold_sql,
"em": em,
"sm": sm,
"exec_acc": exec_acc,
"safety_consistent": safety_consistent,
"latency_ms": latency,
"trace": stages,
"error": None,
}
)
print(f"β
OK | EM={em} | SM={sm} | Exec={exec_acc} | {latency} ms")
except Exception as e:
latency = _int_ms(t0)
results.append(
{
"id": qid,
"question": q,
"sql_pred": None,
"sql_gold": gold_sql,
"em": False,
"sm": False,
"exec_acc": False,
"safety_consistent": None,
"latency_ms": latency,
"trace": [],
"error": str(e),
}
)
print(f"β Fail ({latency} ms): {e}")
time.sleep(args.sleep)
# Summary
total = len(results)
avg_latency = round(sum(r["latency_ms"] for r in results) / max(total, 1), 1)
em_rate = (sum(1 for r in results if r["em"]) / max(total, 1)) if total else 0.0
sm_rate = (sum(1 for r in results if r["sm"]) / max(total, 1)) if total else 0.0
exec_acc_rate = (
(sum(1 for r in results if r["exec_acc"]) / max(total, 1)) if total else 0.0
)
summary: Dict[str, Any] = {
"total": total,
"avg_latency_ms": avg_latency,
"EM": em_rate,
"SM": sm_rate,
"ExecAcc": exec_acc_rate,
"timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
"db": args.db,
"using_default_pipeline": using_default,
}
# Persist outputs (timestamped dir)
out_dir = Path("benchmarks") / "results_pro" / time.strftime("%Y%m%d-%H%M%S")
out_dir.mkdir(parents=True, exist_ok=True)
jsonl_path = out_dir / "spider_eval_pro.jsonl"
with jsonl_path.open("w", encoding="utf-8") as f:
for r in results:
json.dump(r, f, ensure_ascii=False)
f.write("\n")
json_path = out_dir / "summary.json"
with json_path.open("w", encoding="utf-8") as f:
json.dump(summary, f, indent=2)
csv_path = out_dir / "summary.csv"
with csv_path.open("w", newline="", encoding="utf-8") as f:
writer = csv.DictWriter(
f,
fieldnames=["id", "question", "em", "sm", "exec_acc", "latency_ms"],
)
writer.writeheader()
for r in results:
writer.writerow(
{
"id": r["id"],
"question": r["question"],
"em": "β
" if r["em"] else "β",
"sm": "β
" if r["sm"] else "β",
"exec_acc": "β
" if r["exec_acc"] else "β",
"latency_ms": r["latency_ms"],
}
)
print("\nπ Summary:", json.dumps(summary, indent=2))
print(f"πΎ Saved to:\n- {jsonl_path}\n- {json_path}\n- {csv_path}")
if __name__ == "__main__":
main()
|