Spaces:
Sleeping
Sleeping
File size: 10,056 Bytes
af383cf | 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 | """SQLite storage layer for evaluation results.
Why SQLite?
- Zero setup: no database server to run
- File-based: easy to version-control or share
- Fast enough for thousands of eval runs
- Perfect for local development and CI/CD pipelines
Schema design:
- evaluation_runs: One row per EvaluationReport (the "run" metadata)
- evaluation_results: One row per metric score (the individual judge results)
Teaching note: This is a classic "report header + line items" schema pattern.
The header (evaluation_runs) stores summary info, the line items
(evaluation_results) store individual scores. Same pattern used in invoices,
orders, and every reporting system you'll ever build.
"""
import sqlite3
import json
import logging
from datetime import datetime
from pathlib import Path
from typing import Optional
from ..types import (
EvaluationReport,
EvaluationResult,
EvaluationMetric,
)
logger = logging.getLogger(__name__)
CREATE_RUNS_TABLE = """
CREATE TABLE IF NOT EXISTS evaluation_runs (
id TEXT PRIMARY KEY,
system_name TEXT NOT NULL,
timestamp TEXT NOT NULL,
total_examples INTEGER NOT NULL,
summary_scores TEXT NOT NULL, -- JSON: {"faithfulness": 0.87, ...}
judge_models TEXT NOT NULL, -- JSON: ["gpt-4", ...]
total_cost REAL NOT NULL DEFAULT 0.0,
metadata TEXT NOT NULL DEFAULT '{}' -- JSON blob for extra fields
);
"""
CREATE_RESULTS_TABLE = """
CREATE TABLE IF NOT EXISTS evaluation_results (
id INTEGER PRIMARY KEY AUTOINCREMENT,
run_id TEXT NOT NULL REFERENCES evaluation_runs(id),
metric TEXT NOT NULL,
score REAL NOT NULL,
raw_score REAL,
reasoning TEXT NOT NULL,
judge_model TEXT NOT NULL,
confidence REAL NOT NULL DEFAULT 0.5,
timestamp TEXT NOT NULL
);
"""
CREATE_INDEX = """
CREATE INDEX IF NOT EXISTS idx_results_run_id ON evaluation_results(run_id);
CREATE INDEX IF NOT EXISTS idx_runs_system_name ON evaluation_runs(system_name);
CREATE INDEX IF NOT EXISTS idx_runs_timestamp ON evaluation_runs(timestamp);
"""
class EvalDatabase:
"""SQLite-backed storage for evaluation runs and results.
Usage:
db = EvalDatabase("data/results.db")
# Save a report
db.save_report(report)
# Load all runs for a system
runs = db.get_runs_for_system("my-rag-v2")
# Compare two runs
db.compare_runs(run_id_1, run_id_2)
"""
def __init__(self, db_path: str = "data/results.db"):
self.db_path = Path(db_path)
self.db_path.parent.mkdir(parents=True, exist_ok=True)
self._init_db()
def _get_conn(self) -> sqlite3.Connection:
"""Get a database connection with row factory for dict-like access."""
conn = sqlite3.connect(str(self.db_path))
conn.row_factory = sqlite3.Row # Rows accessible as dicts
return conn
def _init_db(self) -> None:
"""Create tables if they don't exist."""
with self._get_conn() as conn:
conn.execute(CREATE_RUNS_TABLE)
conn.execute(CREATE_RESULTS_TABLE)
# CREATE INDEX can't be in a multi-statement execute, so split:
for stmt in CREATE_INDEX.strip().split(";"):
stmt = stmt.strip()
if stmt:
conn.execute(stmt)
logger.debug(f"Database initialized at {self.db_path}")
def save_report(self, report: EvaluationReport) -> None:
"""Persist a full EvaluationReport to the database."""
with self._get_conn() as conn:
# Insert the run header
conn.execute(
"""
INSERT OR REPLACE INTO evaluation_runs
(id, system_name, timestamp, total_examples,
summary_scores, judge_models, total_cost, metadata)
VALUES (?, ?, ?, ?, ?, ?, ?, ?)
""",
(
report.id,
report.system_name,
report.timestamp.isoformat(),
report.total_examples_evaluated,
json.dumps({k.value if hasattr(k, 'value') else k: v
for k, v in report.summary_scores.items()}),
json.dumps(report.judge_models_used),
report.total_cost,
json.dumps(report.metadata),
),
)
# Insert individual results (all metrics, all examples)
for metric, results in report.results.items():
for result in results:
conn.execute(
"""
INSERT INTO evaluation_results
(run_id, metric, score, raw_score, reasoning,
judge_model, confidence, timestamp)
VALUES (?, ?, ?, ?, ?, ?, ?, ?)
""",
(
report.id,
result.metric.value if hasattr(result.metric, 'value') else result.metric,
result.score,
result.raw_score,
result.reasoning,
result.judge_model,
result.confidence,
result.timestamp.isoformat(),
),
)
logger.info(f"Saved report {report.id} for system '{report.system_name}'")
def get_run(self, run_id: str) -> Optional[dict]:
"""Load a single run's metadata by ID."""
with self._get_conn() as conn:
row = conn.execute(
"SELECT * FROM evaluation_runs WHERE id = ?", (run_id,)
).fetchone()
if not row:
return None
result = dict(row)
result["summary_scores"] = json.loads(result["summary_scores"])
result["judge_models"] = json.loads(result["judge_models"])
result["metadata"] = json.loads(result["metadata"])
return result
def get_runs_for_system(self, system_name: str) -> list[dict]:
"""Load all runs for a given system, newest first."""
with self._get_conn() as conn:
rows = conn.execute(
"""
SELECT * FROM evaluation_runs
WHERE system_name = ?
ORDER BY timestamp DESC
""",
(system_name,),
).fetchall()
runs = []
for row in rows:
r = dict(row)
r["summary_scores"] = json.loads(r["summary_scores"])
r["judge_models"] = json.loads(r["judge_models"])
r["metadata"] = json.loads(r["metadata"])
runs.append(r)
return runs
def list_systems(self) -> list[str]:
"""List all system names that have been evaluated."""
with self._get_conn() as conn:
rows = conn.execute(
"SELECT DISTINCT system_name FROM evaluation_runs ORDER BY system_name"
).fetchall()
return [row["system_name"] for row in rows]
def list_all_runs(self, limit: int = 50) -> list[dict]:
"""List recent evaluation runs across all systems."""
with self._get_conn() as conn:
rows = conn.execute(
"""
SELECT * FROM evaluation_runs
ORDER BY timestamp DESC
LIMIT ?
""",
(limit,),
).fetchall()
runs = []
for row in rows:
r = dict(row)
r["summary_scores"] = json.loads(r["summary_scores"])
runs.append(r)
return runs
def get_results_for_run(
self,
run_id: str,
metric: Optional[str] = None,
) -> list[dict]:
"""Load all individual results for a run, optionally filtered by metric."""
query = "SELECT * FROM evaluation_results WHERE run_id = ?"
params: tuple = (run_id,)
if metric:
query += " AND metric = ?"
params = (run_id, metric)
with self._get_conn() as conn:
rows = conn.execute(query, params).fetchall()
return [dict(row) for row in rows]
def compare_runs(self, run_id_1: str, run_id_2: str) -> dict:
"""Compare two runs side-by-side: shows score delta per metric.
Returns a dict: {metric: {run1: score, run2: score, delta: diff}}
"""
run1 = self.get_run(run_id_1)
run2 = self.get_run(run_id_2)
if not run1 or not run2:
raise ValueError("One or both run IDs not found")
comparison = {}
all_metrics = set(run1["summary_scores"]) | set(run2["summary_scores"])
for metric in all_metrics:
s1 = run1["summary_scores"].get(metric, 0.0)
s2 = run2["summary_scores"].get(metric, 0.0)
comparison[metric] = {
"run1_score": s1,
"run2_score": s2,
"delta": s2 - s1, # positive = run2 improved
"improved": s2 > s1,
}
return {
"run1": {"id": run_id_1, "system": run1["system_name"], "timestamp": run1["timestamp"]},
"run2": {"id": run_id_2, "system": run2["system_name"], "timestamp": run2["timestamp"]},
"metrics": comparison,
}
def get_score_history(
self,
system_name: str,
metric: str,
) -> list[dict]:
"""Get the score trend for a metric over time (for dashboard charts)."""
with self._get_conn() as conn:
rows = conn.execute(
"""
SELECT r.timestamp, r.id,
json_extract(r.summary_scores, '$.' || ?) as score
FROM evaluation_runs r
WHERE r.system_name = ?
ORDER BY r.timestamp ASC
""",
(metric, system_name),
).fetchall()
return [{"timestamp": row[0], "run_id": row[1], "score": row[2]} for row in rows]
|