Spaces:
Sleeping
Sleeping
File size: 7,533 Bytes
210535c 429a3ac 210535c 429a3ac 210535c 429a3ac 210535c 429a3ac 210535c | 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 | """
Core OpenEnv environment: SQLOptimizerEnv
Implements the three required methods:
reset(task_id) → Observation
step(action) → (Observation, Reward, done, info)
state() → dict (current internal snapshot)
"""
from __future__ import annotations
from typing import Any, Dict, Optional, Tuple
from .models import Action, Observation, Reward, RewardBreakdown
from .tasks import TASKS, TaskDef, get_task
from .reward import compute_step_reward
_MIN_SCORE_EPS = 0.001
_MAX_SCORE_EPS = 0.999
def _strict_score(value: float) -> float:
return round(min(max(float(value), _MIN_SCORE_EPS), _MAX_SCORE_EPS), 4)
class SQLOptimizerEnv:
"""SQL Query Optimizer OpenEnv environment."""
def __init__(self) -> None:
self._task: Optional[TaskDef] = None
self._step_number: int = 0
self._done: bool = False
self._cumulative_score: float = 0.0
self._prev_grader_score: float = 0.0
self._history: list[Dict[str, Any]] = []
self._last_grader_score: float = 0.0
# ──────────────────────────────────────────────────────────────────────────
# reset
# ──────────────────────────────────────────────────────────────────────────
def reset(self, task_id: int = 1) -> Observation:
"""Start a fresh episode for the given task."""
self._task = get_task(task_id)
self._step_number = 0
self._done = False
self._cumulative_score = 0.0
self._prev_grader_score = 0.0
self._last_grader_score = 0.0
self._history = []
return self._make_observation()
# ──────────────────────────────────────────────────────────────────────────
# step
# ──────────────────────────────────────────────────────────────────────────
def step(self, action: Action) -> Tuple[Observation, Reward, bool, Dict[str, Any]]:
"""
Advance the environment by one step.
Returns:
observation: next Observation
reward: Reward for this step
done: whether the episode has ended
info: auxiliary dict
"""
if self._task is None:
raise RuntimeError("Call reset() before step().")
if self._done:
raise RuntimeError("Episode is done. Call reset() to start a new episode.")
# Validate action
is_invalid = not action.rewritten_query or not action.rewritten_query.strip()
# Run grader
if is_invalid:
grader_result_score = self._prev_grader_score
breakdown = RewardBreakdown()
feedback = "Empty or invalid query submitted."
else:
gr = self._task.grader(action.rewritten_query)
grader_result_score = gr.score
breakdown = RewardBreakdown(
correctness=gr.correctness,
performance=gr.performance,
style=gr.style,
step_penalty=0.0,
)
feedback = gr.feedback
grader_result_score = _strict_score(grader_result_score)
# Compute shaped reward
step_reward = compute_step_reward(
grader_score=grader_result_score,
prev_grader_score=self._prev_grader_score,
step_number=self._step_number,
max_steps=self._task.max_steps,
is_done=action.is_done,
is_invalid=is_invalid,
)
# Apply step penalty to breakdown
import math
halfway = math.ceil(self._task.max_steps / 2)
if self._step_number > halfway and not action.is_done:
breakdown.step_penalty = -0.02
self._cumulative_score = _strict_score(self._cumulative_score + step_reward)
self._prev_grader_score = grader_result_score
self._last_grader_score = grader_result_score
self._step_number += 1
# Episode ends if agent signals done OR max steps reached
self._done = action.is_done or self._step_number >= self._task.max_steps
# Record history
self._history.append(
{
"step": self._step_number,
"rewritten_query": action.rewritten_query,
"grader_score": grader_result_score,
"step_reward": step_reward,
"is_done": action.is_done,
}
)
reward = Reward(
score=_strict_score(step_reward),
grader_score=grader_result_score,
breakdown=breakdown,
feedback=feedback,
cumulative_score=self._cumulative_score,
)
info = {
"step_number": self._step_number,
"grader_score": grader_result_score,
"cumulative_score": self._cumulative_score,
"is_invalid": is_invalid,
}
return self._make_observation(), reward, self._done, info
# ──────────────────────────────────────────────────────────────────────────
# state
# ──────────────────────────────────────────────────────────────────────────
def state(self) -> Dict[str, Any]:
"""Return the current internal state snapshot."""
if self._task is None:
return {"status": "not_started"}
return {
"task_id": self._task.id,
"task_name": self._task.name,
"difficulty": self._task.difficulty,
"step_number": self._step_number,
"max_steps": self._task.max_steps,
"done": self._done,
"cumulative_score": self._cumulative_score,
"last_grader_score": self._last_grader_score,
"history": self._history,
}
# ──────────────────────────────────────────────────────────────────────────
# Internal helpers
# ──────────────────────────────────────────────────────────────────────────
def _make_observation(self) -> Observation:
assert self._task is not None
return Observation(
task_id=self._task.id,
task_name=self._task.name,
task_description=self._task.description,
query=self._task.query,
schema_context=self._task.schema_context,
hint=self._task.hint,
step_number=self._step_number,
max_steps=self._task.max_steps,
done=self._done,
)
|