Apppublic

dkAmulet/sql-query-optimizer

sourceHugging Facemitupdated 5mo agoView on Hugging Face

0likes

App README

SQL Query Optimizer — OpenEnv Environment

An RL environment where AI agents learn to rewrite slow SQL queries for correctness, performance, and style — against a live SQLite e-commerce database.

![OpenEnv]() ![Python 3.11]() ![License MIT]()

Quick Start

bash

# Reset to the easiest task
curl -X POST "$SPACE_URL/reset?task_id=select_star_removal"

# Submit an optimised query
curl -X POST "$SPACE_URL/step" \
     -H "Content-Type: application/json" \
     -d '{"optimized_query": "SELECT user_id, username, email FROM users WHERE is_active = 1"}'

# Check current episode state
curl "$SPACE_URL/state"

Interactive API docs: $SPACE_URL/docs

Tasks

ID	Name	Difficulty	Max Steps
`select_star_removal`	SELECT * Elimination	Easy	5
`subquery_to_join`	Correlated Subquery → JOIN	Medium	6
`aggregation_optimization`	Aggregation Optimization	Hard	8

Reward Dimensions

Every step returns a dense, partial-credit reward:

Dimension	Max	Signal
`validity`	0.10	Query runs without error
`correctness`	0.35–0.40	Same result set as reference
`performance`	0.20–0.30	Index usage in EXPLAIN plan
`style`	0.20–0.30	No SELECT *, no correlated subqueries

Baseline Scores (GPT-4o-mini)

Task	Score
SELECT * Elimination	0.70
Correlated Subquery → JOIN	0.60
Aggregation Optimization	0.45
Overall	0.58

Full documentation in README.md.