fixed

data_cleaning_env.py

@@ -214,6 +214,16 @@ class DataCleaningEnvironment:
             tasks_completed=[],
             total_reward=self._total_reward
         )
+    
+    @staticmethod
+    def _parse_date(val):
+        for fmt in ["%Y-%m-%d", "%Y/%m/%d", "%Y.%m.%d",
+                "%m/%d/%Y", "%m-%d-%Y", "%d/%m/%Y", "%d-%m-%Y", "%d.%m.%Y"]:
+            try:
+                return datetime.strptime(str(val).strip(), fmt).strftime("%Y-%m-%d")
+            except:
+                continue
+        return None
 
     def _apply_action(self, action: CleaningAction) -> pd.DataFrame:
         df = deepcopy(self._current_df)
@@ -223,7 +233,7 @@ class DataCleaningEnvironment:
             col = action.column or "hire_date"
             if col not in df.columns:
                 raise ValueError(f"Column '{col}' not found.")
-            df[col] = pd.to_datetime(df[col], errors='coerce', dayfirst=True).dt.strftime("%Y-%m-%d")
+            df[col] = df[col].apply(self._parse_date)
         elif action.action_type == "remove_outliers":
             target_cols = [action.column] if action.column and action.column != "all" else ['salary', 'age']
             for col in target_cols:

inference.py

@@ -1,38 +1,28 @@
 """
-inference.py — Mandatory OpenEnv submission file
-=================================================
-Runs all 3 tasks and produces reproducible scores.
-
-Usage:
-    export API_BASE_URL=https://router.huggingface.co/v1
-    export MODEL_NAME=SohamK18/data-cleaning-grpo
-    export HF_TOKEN=hf_your_token_here
-    export ENV_URL=https://your-space.hf.space   # your HF Space URL
-    python inference.py
+inference.py — OpenEnv submission file
 """
-
 import os, json, sys
 from openai import OpenAI
 from data_cleaning_env import DataCleaningEnvironment, CleaningAction
 
-# ── Config ──────────────────────────────────────────────────────
+# ── Config ────────────────────────────────────────────────────────────────────
 API_BASE_URL = os.getenv("API_BASE_URL", "https://router.huggingface.co/v1")
-HF_TOKEN     = os.getenv("HF_TOKEN") or os.getenv("API_KEY", "")
-MODEL_NAME   = os.getenv("MODEL_NAME", "SohamK18/data-cleaning-grpo")
-N_EPISODES   = int(os.getenv("N_EPISODES", "10"))
-
-if not HF_TOKEN:
-    print("ERROR: HF_TOKEN not set.", file=sys.stderr)
-    sys.exit(1)
+HF_TOKEN     = os.getenv("HF_TOKEN")
+MODEL_NAME = os.getenv("MODEL_NAME", "Qwen/Qwen2.5-1.5B-Instruct")
+if HF_TOKEN is None:
+    raise ValueError("HF_TOKEN environment variable is required")
 
 client = OpenAI(base_url=API_BASE_URL, api_key=HF_TOKEN)
 
 SYSTEM_PROMPT = (
     "You are a data cleaning expert. "
     "Respond ONLY with a valid JSON object, no markdown, no explanation.\n"
-    "Format: {\"action_type\": \"<remove_nulls|fix_dates|remove_outliers>\", \"column\": \"<col_or_null>\"}"
+    'Format: {"action_type": "<remove_nulls|fix_dates|remove_outliers>", "column": "<col_or_null>"}'
 )
 
+TASK_NAMES = {1: "remove_nulls", 2: "fix_dates", 3: "remove_outliers"}
+ENV_NAME   = "data_cleaning"
+
 
 def parse_llm_response(text: str, task_id: int) -> CleaningAction:
     text = text.strip().replace("```json", "").replace("```", "").strip()
@@ -41,7 +31,11 @@ def parse_llm_response(text: str, task_id: int) -> CleaningAction:
         action_type = data.get("action_type", "remove_nulls")
         if action_type not in ["remove_nulls", "fix_dates", "remove_outliers"]:
             action_type = "remove_nulls"
-        return CleaningAction(task_id=task_id, action_type=action_type, column=data.get("column"))
+        return CleaningAction(
+            task_id=task_id,
+            action_type=action_type,
+            column=data.get("column")
+        )
     except Exception:
         if "date" in text.lower():
             return CleaningAction(task_id=task_id, action_type="fix_dates", column="hire_date")
@@ -50,9 +44,20 @@ def parse_llm_response(text: str, task_id: int) -> CleaningAction:
         return CleaningAction(task_id=task_id, action_type="remove_nulls")
 
 
-def run_episode(task_id: int, seed: int) -> float:
+def heuristic_action(task_id: int, obs) -> CleaningAction:
+    if obs.null_count > 0:
+        return CleaningAction(task_id=task_id, action_type="remove_nulls")
+    elif obs.date_format_errors > 0:
+        return CleaningAction(task_id=task_id, action_type="fix_dates", column="hire_date")
+    else:
+        return CleaningAction(task_id=task_id, action_type="remove_outliers", column="all")
+
+
+def run_episode(task_id: int, seed: int):
     env = DataCleaningEnvironment(task_id=task_id, seed=seed)
     obs = env.reset()
+    error_str = "null"
+    action = None
 
     user_msg = (
         f"Task {task_id}: {obs.task_description}\n"
@@ -62,66 +67,78 @@ def run_episode(task_id: int, seed: int) -> float:
         f"Respond with JSON only."
     )
 
+    # ── Primary: LLM via OpenAI client ───────────────────────────────────────
     try:
         resp = client.chat.completions.create(
             model=MODEL_NAME,
             messages=[
                 {"role": "system", "content": SYSTEM_PROMPT},
-                {"role": "user",   "content": user_msg}
+                {"role": "user",   "content": user_msg},
             ],
             max_tokens=100,
-            temperature=0.1
+            temperature=0.1,
         )
-        text = resp.choices[0].message.content
-        action = parse_llm_response(text, task_id)
-    except Exception:
-        # Heuristic fallback
-        if obs.null_count > 0:
-            action = CleaningAction(task_id=task_id, action_type="remove_nulls")
-        elif obs.date_format_errors > 0:
-            action = CleaningAction(task_id=task_id, action_type="fix_dates", column="hire_date")
-        else:
-            action = CleaningAction(task_id=task_id, action_type="remove_outliers", column="all")
+        action = parse_llm_response(resp.choices[0].message.content, task_id)
+    except Exception as e:
+        error_str = str(e).replace("\n", " ")
 
-    _, reward, _, _ = env.step(action)
-    return float(reward)
+    # ── Fallback: heuristic if LLM failed ────────────────────────────────────
+    if action is None:
+        action = heuristic_action(task_id, obs)
 
+    col = action.column if action.column else "null"
+    action_str = f"{action.action_type}('{col}')"
 
-def main():
-    TASK_NAMES = {1: "remove_nulls", 2: "fix_dates", 3: "remove_outliers"}
-    DIFFICULTIES = {1: "easy", 2: "medium", 3: "hard"}
+    _, reward, done, _ = env.step(action)
+    if hasattr(env, "close"):
+        env.close()
+
+    return float(reward), action_str, bool(done), error_str
 
+
+def main():
     all_results = {}
+    n_episodes = int(os.getenv("N_EPISODES", "10"))
 
     for task_id in [1, 2, 3]:
         task_name = TASK_NAMES[task_id]
-        difficulty = DIFFICULTIES[task_id]
-
-        # ── [START] log ──────────────────────────────────────────
-        print(f"[START] task={task_name}", flush=True)
-
-        episode_scores = []
-        for seed in range(N_EPISODES):
-            reward = run_episode(task_id, seed)
-            episode_scores.append(reward)
-
-            # ── [STEP] log ────────────────────────────────────────
-            print(f"[STEP] step={seed+1} reward={reward}", flush=True)
-
-        avg_score = round(sum(episode_scores) / len(episode_scores), 4)
-        all_results[task_id] = avg_score
-
-        # ── [END] log ─────────────────────────────────────────────
-        print(f"[END] task={task_name} score={avg_score} steps={len(episode_scores)}", flush=True)
-
-    overall = round(sum(all_results.values()) / 3, 4)
-
-    # Write scores to file for automated evaluation
+        print(f"[START] task={task_name} env={ENV_NAME} model={MODEL_NAME}", flush=True)
+
+        episode_rewards = []
+        success = False
+        score = 0.0
+
+        try:
+            for seed in range(n_episodes):
+                reward, action_str, done, error_str = run_episode(task_id, seed)
+                episode_rewards.append(reward)
+                print(
+                    f"[STEP] step={seed + 1} action={action_str} "
+                    f"reward={reward:.2f} done={str(done).lower()} error={error_str}",
+                    flush=True,
+                )
+
+            score = sum(episode_rewards) / len(episode_rewards)
+            score = round(min(max(score, 0.0), 1.0), 2)
+            all_results[task_id] = score
+            success = score > 0.0
+
+        finally:
+            rewards_str = ",".join(f"{r:.2f}" for r in episode_rewards)
+            # ── [END] with score= field as required ──────────────────────────
+            print(
+                f"[END] success={str(success).lower()} "
+                f"steps={len(episode_rewards)} "
+                f"score={score:.2f} "
+                f"rewards={rewards_str}",
+                flush=True,
+            )
+
+    overall = round(sum(all_results.values()) / max(len(all_results), 1), 4)
     with open("scores.json", "w") as f:
         json.dump({"tasks": all_results, "overall": overall}, f, indent=2)
-
     print(f"[SUMMARY] overall_score={overall} task_scores={all_results}", flush=True)
 
 
 if __name__ == "__main__":
-    main()
+    main()