Remove strip from ProceduralDataset::core score_answer()

reasoning_gym/arithmetic/chain_sum.py

@@ -111,7 +111,8 @@ def _generate_task(self, rng: random.Random, num_terms: int, min_value: int, max
         return expression, result
 
     def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
-        return utils.compute_decimal_reward(answer, oracle_answer=entry["answer"])
+        # tolerate sign, leading zeros and trailing decimals, strip commas "+01,000.00" == "1000"
+        return utils.compute_decimal_reward(answer, oracle_answer=entry["answer"], strip_commas=True)
 
 
 class ChainSumCurriculum(BaseCurriculum):

reasoning_gym/arithmetic/products.py

@@ -103,7 +103,8 @@ def _generate_task(self, rng: random.Random, num_terms: int, min_value: int, max
         return expression, result
 
     def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
-        return utils.compute_decimal_reward(answer, oracle_answer=entry["answer"])
+        # tolerate sign, leading zeros and trailing decimals, strip commas "+01,000.00" == "1000"
+        return utils.compute_decimal_reward(answer, oracle_answer=entry["answer"], strip_commas=True)
 
 
 class ProductsCurriculum(BaseCurriculum):

reasoning_gym/dataset.py

@@ -62,10 +62,9 @@ def __getitem__(self, idx: int) -> dict[str, Any]:
 
     def score_answer(self, answer: Optional[str], entry: dict[str, Any]) -> float:
         """Overwrite this method in derived classes if a single oracle answer is not available."""
-        oracle_answer = entry["answer"].strip()
+        oracle_answer = entry["answer"]
         reward = 0.0
-        if answer is not None and len(answer) > 0:
-            answer = answer.strip()
+        if isinstance(answer, str) and len(answer) > 0:
             if answer == oracle_answer:
                 reward = 1.0
             elif oracle_answer in answer:

reasoning_gym/utils.py

@@ -19,7 +19,7 @@
 }
 
 
-def extract_answer(completion: str, tag_name: str = "answer") -> Optional[str]:
+def extract_answer(completion: str, tag_name: str = "answer", strip: bool = True) -> Optional[str]:
     regex = f"<{tag_name}>\\s?(.*?)\\s?</{tag_name}>"
     matches = list(
         re.finditer(
@@ -30,21 +30,25 @@ def extract_answer(completion: str, tag_name: str = "answer") -> Optional[str]:
     )
     if not matches:
         return None
-    return matches[-1].group(1)
+    answer = matches[-1].group(1)
+    if strip:
+        answer = answer.strip()
+    return answer
 
 
-def format_number(num: Union[int, float], max_decimals: int = 2) -> str:
+def format_number(num: Union[int, float], max_decimals: int = 2, round_if_needed: bool = False) -> str:
     """Convert a number to string representation with controlled decimal places.
 
     Args:
         num: Number to format
         max_decimals: Maximum allowed decimal places
+        round_if_needed: If True, round the number to max_decimals instead of raising an error
 
     Returns:
         String representation of the number
 
     Raises:
-        ValueError: If number requires more decimal places than allowed
+        ValueError: If number requires more decimal places than allowed and round_if_needed is False
     """
     if isinstance(num, int) or num.is_integer():
         return str(int(num))
@@ -57,19 +61,20 @@ def format_number(num: Union[int, float], max_decimals: int = 2) -> str:
     str_val = str_val.rstrip("0").rstrip(".")
     if "." in str_val:
         required_decimals = len(str_val.split(".")[1])
-        if required_decimals > max_decimals:
+        if required_decimals > max_decimals and not round_if_needed:
             raise ValueError(f"Number {num} requires {required_decimals} decimals but only {max_decimals} allowed")
 
-    # Format with required decimals
+    # Format with required decimals (will round if needed)
     result = f"{num:.{max_decimals}f}".rstrip("0").rstrip(".")
 
-    # Verify result parses back to original value
-    try:
-        parsed = float(result)
-        if not math.isclose(parsed, num, rel_tol=1e-9):
-            raise ValueError(f"String representation {result} does not match original value {num}")
-    except (ValueError, InvalidOperation) as e:
-        raise ValueError(f"Failed to verify string representation: {e}")
+    # Verify result parses back to original value (skip verification if rounding was applied)
+    if not round_if_needed:
+        try:
+            parsed = float(result)
+            if not math.isclose(parsed, num, rel_tol=1e-9):
+                raise ValueError(f"String representation {result} does not match original value {num}")
+        except (ValueError, InvalidOperation) as e:
+            raise ValueError(f"Failed to verify string representation: {e}")
 
     return result
 
@@ -84,6 +89,7 @@ def is_integer(obj: Any) -> bool:
 
 def compute_decimal_reward(answer: Optional[str], oracle_answer: str, strip_commas: bool = True) -> float:
     """Compute the reward for a given answer compared to the oracle answer.
+    Tolerate sign, leading zeros and trailing decimals, optionally strip commas ("+01,000.00" == "1000")
 
     Args:
         answer: Answer provided by model
@@ -102,7 +108,7 @@ def compute_decimal_reward(answer: Optional[str], oracle_answer: str, strip_comm
                 oracle_answer = oracle_answer.replace(",", "")
 
             if Decimal(answer) == Decimal(oracle_answer):
-                reward = 1.0
+                return 1.0
         except:
             pass
 

tests/test_dataset.py

@@ -2,7 +2,6 @@
 
 from reasoning_gym.arithmetic.basic_arithmetic import BasicArithmeticDataset, BasicArithmeticDatasetConfig
 from reasoning_gym.dataset import ReseedingDataset
-from reasoning_gym.utils import extract_answer
 
 
 def test_reseeding_dataset_iteration():
@@ -41,12 +40,7 @@ def test_reseeding_dataset_iteration():
     assert infinite_dataset.score_answer(test_item["answer"], test_item) == 1.0
 
 
-def test_extract_answer():
-    assert extract_answer("This is a text. <final_answer>1234</final_answer>", tag_name="final_answer") == "1234"
-
-    # ignore single whitespae
-    assert extract_answer("This is a text. <answer>\n1234 </answer>", tag_name="answer") == "1234"
-
+def test_basic_arithmetic_score_answer():
     config = BasicArithmeticDatasetConfig(
         min_terms=2, max_terms=3, min_digits=1, max_digits=2, operators=["+"], allow_parentheses=False, seed=42, size=10
     )

tests/test_utils.py

@@ -0,0 +1,53 @@
+import pytest
+
+from reasoning_gym.utils import compute_decimal_reward, extract_answer, format_number
+
+
+def test_extract_answer():
+    assert extract_answer("This is a text. <final_answer>1234</final_answer>", tag_name="final_answer") == "1234"
+
+    # ignore whitespaces
+    assert extract_answer("This is a text. <answer>\n1234 </answer>", tag_name="answer", strip=True) == "1234"
+
+
+def test_format_number():
+    # Test integers
+    assert format_number(42) == "42"
+    assert format_number(42.0) == "42"
+
+    # Test decimals
+    assert format_number(3.14) == "3.14"
+    assert format_number(3.10) == "3.1"
+    assert format_number(3.00) == "3"
+
+    # Test with max_decimals (rounding)
+    assert format_number(3.14159, max_decimals=4, round_if_needed=True) == "3.1416"
+
+    # Test with trailing zeros
+    assert format_number(5.5000) == "5.5"
+
+    # Test error cases
+    with pytest.raises(ValueError):
+        format_number(3.14159, max_decimals=2)
+
+
+def test_compute_decimal_reward():
+    # Test exact matches
+    assert compute_decimal_reward("42", "42") == 1.0
+    assert compute_decimal_reward("3.14", "3.14") == 1.0
+
+    # Test with commas
+    assert compute_decimal_reward("1,000", "1000") == 1.0
+    assert compute_decimal_reward("1,000", "1000", strip_commas=False) < 1.0
+
+    # Test with sign, leading zeros, and trailing decimals
+    assert compute_decimal_reward("+0001,000.00", "1000") == 1.0
+
+    # Test partial matches
+    assert compute_decimal_reward("The answer is 42", "42") < 1.0
+    assert compute_decimal_reward("The answer is 42", "42") > 0.01
+
+    # Test invalid answers
+    assert compute_decimal_reward(None, "42") == 0.0
+    assert compute_decimal_reward("", "42") == 0.0
+    assert compute_decimal_reward("not a number", "42") == 0.01