File size: 6,660 Bytes
72805b8 ac49ad8 72805b8 | 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 | """
Core SQL Arena Environment.
Implements the OpenEnv step()/reset()/state() interface.
"""
from typing import Optional, Dict, Any, List
from .models import SQLArenaAction, SQLArenaObservation, SQLArenaState
from .database import DatabaseManager
from .tasks import SQLTask, get_task, list_tasks, TASK_BY_ID
from .graders import grade_result, generate_hint
class StepResult:
"""Result of a single environment step."""
def __init__(
self,
observation: SQLArenaObservation,
reward: float,
done: bool,
info: Optional[Dict[str, Any]] = None,
):
self.observation = observation
self.reward = reward
self.done = done
self.info = info or {}
class SQLArenaEnvironment:
"""
SQL Arena: An interactive SQL query challenge environment.
The agent receives a database schema and a natural language question,
then iteratively writes SQL queries. The environment provides
execution results, feedback, and partial credit scoring.
"""
def __init__(self):
self.db = DatabaseManager()
self.current_task: Optional[SQLTask] = None
self._state: Optional[SQLArenaState] = None
self._last_observation: Optional[SQLArenaObservation] = None
def reset(
self,
difficulty: str = "basic_select",
task_id: Optional[str] = None,
) -> StepResult:
"""
Reset the environment with a new task.
Args:
difficulty: 'basic_select', 'join_aggregate', or 'complex_analysis'
task_id: Optional specific task ID
Returns:
StepResult with initial observation
"""
# Get the task
self.current_task = get_task(difficulty, task_id)
task = self.current_task
# Setup database
self.db.create_database(task.setup_sql)
# Initialize state
self._state = SQLArenaState(
task_id=task.task_id,
difficulty=task.difficulty,
current_step=0,
max_steps=task.max_steps,
best_score=0.0,
total_reward=0.0,
rewards_history=[],
done=False,
last_action_error=None,
)
# Create initial observation
self._last_observation = SQLArenaObservation(
schema_description=task.schema_description,
question=task.question,
query_result=None,
error_message=None,
feedback="Welcome to SQL Arena! Write a SQL query to answer the question above.",
expected_columns=task.expected_columns,
attempts_remaining=task.max_steps,
difficulty=task.difficulty,
task_id=task.task_id,
)
return StepResult(
observation=self._last_observation,
reward=0.0,
done=False,
info={"task_title": task.title},
)
def step(self, action: SQLArenaAction) -> StepResult:
"""
Execute the agent's SQL query and return feedback.
Args:
action: SQLArenaAction containing the SQL query
Returns:
StepResult with observation, reward, and done flag
"""
if self._state is None or self.current_task is None:
raise RuntimeError("Environment not initialized. Call reset() first.")
if self._state.done:
raise RuntimeError("Episode is done. Call reset() to start a new episode.")
task = self.current_task
state = self._state
# Increment step counter
state.current_step += 1
# Execute the query
success, result, error = self.db.execute_query(action.sql_query)
# Grade the result
score, feedback = grade_result(task, success, result, error)
# Track best score
state.best_score = max(state.best_score, score)
# Calculate step reward
if len(state.rewards_history) == 0:
reward = score
else:
prev_best = max(state.rewards_history) if state.rewards_history else 0.0
improvement = max(0, score - prev_best)
reward = score * 0.5 + improvement * 0.5
reward = round(min(max(reward, 0.0), 1.0), 4)
# Clamp to strictly between 0 and 1
if reward <= 0.0:
reward = 0.01
if reward >= 1.0:
reward = 0.99
state.rewards_history.append(reward)
state.total_reward += reward
# Add progressive hints
hint = generate_hint(task, state.current_step, score)
if hint and score < 1.0:
feedback += f"\n\n{hint}"
# Check if done
attempts_remaining = task.max_steps - state.current_step
is_perfect = score >= 1.0
is_out_of_steps = attempts_remaining <= 0
state.done = is_perfect or is_out_of_steps
state.last_action_error = error
# Format query result for observation
query_result_str = None
if success and result:
query_result_str = self.db.format_result(result)
# Build observation
self._last_observation = SQLArenaObservation(
schema_description=task.schema_description,
question=task.question,
query_result=query_result_str,
error_message=error,
feedback=feedback,
expected_columns=task.expected_columns,
attempts_remaining=attempts_remaining,
difficulty=task.difficulty,
task_id=task.task_id,
)
return StepResult(
observation=self._last_observation,
reward=reward,
done=state.done,
info={
"score": score,
"best_score": state.best_score,
"step": state.current_step,
"is_perfect": is_perfect,
},
)
def state(self) -> SQLArenaState:
"""Return the current environment state."""
if self._state is None:
raise RuntimeError("Environment not initialized. Call reset() first.")
return self._state
def close(self) -> None:
"""Clean up resources."""
self.db.close()
self.current_task = None
self._state = None
self._last_observation = None
def get_available_tasks(self) -> Dict:
"""Return all available tasks."""
return list_tasks() |