Spaces:
Sleeping
Sleeping
File size: 6,151 Bytes
269f632 | 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 | """SQLite-backed environment state for SQL repair tasks.
The env exposes a minimal Gym-like API:
reset(task_id) -> observation dict
step(action) -> {observation, reward, done, info}
Per-task state is held in this single instance for simplicity. The
validator only needs one parallel run.
"""
from __future__ import annotations
import sqlite3
from typing import Any, Dict, List, Optional
from .tasks import TASKS, TASK_IDS
MAX_STEPS = 6
def _new_db(task_id: str) -> sqlite3.Connection:
"""Build a fresh in-memory DB for the given task."""
if task_id not in TASKS:
raise KeyError(f"Unknown task_id: {task_id}")
conn = sqlite3.connect(":memory:")
cur = conn.cursor()
for stmt in TASKS[task_id]["schema"]:
cur.execute(stmt)
conn.commit()
return conn
def _run_query(task_id: str, query: str) -> Dict[str, Any]:
"""Execute a query against a fresh DB; return rows or error info."""
conn = _new_db(task_id)
try:
cur = conn.execute(query)
rows = cur.fetchall()
col_names = [d[0] for d in cur.description] if cur.description else []
return {"ok": True, "rows": rows, "columns": col_names, "error": None}
except Exception as exc:
return {"ok": False, "rows": None, "columns": [], "error": str(exc)}
finally:
conn.close()
def _expected_rows(task_id: str) -> List[tuple]:
"""Compute the canonical (expected) result set for a task."""
res = _run_query(task_id, TASKS[task_id]["canonical_query"])
if not res["ok"]:
# Should never happen — canonical queries are vetted in tests.
raise RuntimeError(
f"Canonical query for {task_id} failed: {res['error']}"
)
return res["rows"]
class EnvState:
"""Mutable per-session env state. One instance handles all tasks."""
def __init__(self) -> None:
self.task_id: Optional[str] = None
self.step_count: int = 0
self.last_query: Optional[str] = None
self.last_error: Optional[str] = None
self.last_result: Optional[List[tuple]] = None
self.solved: bool = False
self.expected_rows: List[tuple] = []
self.expected_columns: int = 0
# ------------------------------------------------------------------
def reset(self, task_id: Optional[str] = None) -> Dict[str, Any]:
tid = task_id or "task_1"
if tid not in TASKS:
tid = "task_1"
task = TASKS[tid]
self.task_id = tid
self.step_count = 0
self.last_query = None
self.last_error = None
self.last_result = None
self.solved = False
self.expected_rows = _expected_rows(tid)
self.expected_columns = (
len(self.expected_rows[0]) if self.expected_rows else 0
)
# Surface what the broken query actually does, so the agent has
# an error message and a canonical "what went wrong" hint.
baseline = _run_query(tid, task["broken_query"])
return {
"task_id": tid,
"name": task["name"],
"difficulty": task["difficulty"],
"schema_sql": "\n".join(task["schema"]),
"broken_query": task["broken_query"],
"broken_query_error": baseline["error"],
"broken_query_executes": baseline["ok"],
"hint": task["hint"],
"expected_row_count": len(self.expected_rows),
"expected_column_count": self.expected_columns,
"step_count": 0,
"max_steps": MAX_STEPS,
"remaining_steps": MAX_STEPS,
}
# ------------------------------------------------------------------
def step(self, action: Dict[str, Any]) -> Dict[str, Any]:
if self.task_id is None:
return {
"observation": {"error": "No active task. Call /reset first."},
"reward": 0.0,
"done": True,
"info": {"solved": False, "no_active_task": True},
}
self.step_count += 1
action_type = (action or {}).get("action_type", "submit_query")
query = ((action or {}).get("query") or "").strip()
self.last_query = query
reward = 0.0
result_rows: Optional[List[tuple]] = None
error: Optional[str] = None
if action_type != "submit_query":
error = f"Unsupported action_type: {action_type}"
reward = -0.05
elif not query:
error = "Empty query string."
reward = -0.05
else:
res = _run_query(self.task_id, query)
if res["ok"]:
result_rows = res["rows"]
self.last_result = result_rows
self.last_error = None
if result_rows == self.expected_rows:
reward = 1.0
self.solved = True
else:
# executed but wrong rows — small positive reward
reward = 0.4
else:
error = res["error"]
self.last_error = error
self.last_result = None
reward = -0.10
done = self.solved or self.step_count >= MAX_STEPS
observation = {
"task_id": self.task_id,
"step_count": self.step_count,
"submitted_query": query,
"error": error,
"executed": error is None and result_rows is not None,
"matches_expected": (
result_rows == self.expected_rows if result_rows is not None else False
),
"result_row_count": len(result_rows) if result_rows is not None else 0,
"expected_row_count": len(self.expected_rows),
"result_preview": result_rows[:3] if result_rows else None,
"expected_preview": self.expected_rows[:3],
"remaining_steps": max(0, MAX_STEPS - self.step_count),
}
return {
"observation": observation,
"reward": float(reward),
"done": bool(done),
"info": {"solved": self.solved},
}
|