Fixed countdown score_answer (#265)

* fixed countdown score ans
* checked solution uses all numbers

reasoning_gym/games/countdown.py

@@ -1,3 +1,4 @@
+import re
 from dataclasses import dataclass
 from random import Random
 from typing import Any, Optional
@@ -51,9 +52,9 @@ class CountdownDataset(ProceduralDataset):
 
     def __init__(self, config: CountdownConfig):
         self._prompt_templates = [
-            "Using the numbers {numbers}, create an expression that equals {target}.\nYou can only use each number once.",
-            "Find a way to make {target} using some or all of these numbers: {numbers}.\nEach number can only be used once.",
-            "Calculate {target} using the numbers {numbers}.\nEach number may be used at most once.",
+            "Using all the numbers {numbers}, create an expression that equals {target}.\nYou can only use each number once.",
+            "Find a way to make {target} using all of these numbers: {numbers}.\nEach number can only be used once.",
+            "Calculate {target} using all of these numbers: {numbers}.\nEach number may be used at most once.",
         ]
         super().__init__(config=config, seed=config.seed, size=config.size)
 
@@ -174,21 +175,23 @@ def _generate_expression(self, rng: Random) -> tuple[str, list[int], int]:
 
     def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
         """Determine if the solution provided solves the problem"""
-        reward = 0.0
-        metadata = entry["metadata"]
-        if answer is not None:
-            try:
-                user_answer = int(parse_expr(answer))
-                solved = user_answer == metadata["target"]
-                if solved:
-                    reward = 1.0
-                elif len(answer.strip()) > 0:  # encourage partial solutions
-                    reward = 0.05
-                else:
-                    reward = 0.01
-            except:
-                reward = 0.01
-        return reward
+        reward = 0.01  # Default reward
+
+        if answer is None or not answer.strip():
+            return reward
+
+        try:
+            answer = answer.strip()
+            user_answer = int(parse_expr(answer))
+            used_numbers = [int(num) for num in re.findall(r"\b\d+\b", answer)]
+            target_numbers = set(entry["metadata"]["numbers"])
+
+            if (user_answer == entry["metadata"]["target"]) and (set(used_numbers) == target_numbers):
+                return 1.0
+
+            return 0.05 if answer else 0.01
+        except Exception:
+            return 0.01
 
 
 # Register the dataset

tests/test_countdown.py

@@ -72,7 +72,7 @@ def test_countdown_game_items():
             dataset.score_answer(answer="a wrong solution", entry=item) == 0.01
         )  # wrong answer but incorrectly formatted
         assert dataset.score_answer(answer="", entry=item) == 0.01  # wrong answer but empty string
-        assert dataset.score_answer(answer=None, entry=item) == 0.0  # no answer
+        assert dataset.score_answer(answer=None, entry=item) == 0.01  # no answer
 
         try:
             result = eval(expr)  # Safe here since we control expression generation
@@ -81,6 +81,25 @@ def test_countdown_game_items():
             pytest.fail(f"Invalid expression generated: {expr}")
 
 
+def test_answer_with_incorrect_numbers():
+    dataset = CountdownDataset(CountdownConfig(size=10, seed=42))
+    answer = "45+2"
+    item = {
+        "metadata": {
+            "numbers": [44, 3],
+            "target": 47,
+        }
+    }
+    assert dataset.score_answer(answer=answer, entry=item) == 0.05
+
+
+def test_answer_without_all_numbers():
+    dataset = CountdownDataset(CountdownConfig(size=10, seed=42))
+    answer = "45+2+3"
+    item = {"metadata": {"numbers": [1, 45, 2, 3], "target": 50}}
+    assert dataset.score_answer(answer=answer, entry=item) == 0.05
+
+
 def test_countdown_game_randomization():
     """Test number randomization configuration"""
     config = CountdownConfig(min_numbers=4, max_numbers=4, shuffle=False, size=10, seed=42)  # Fixed size for testing