🧪 Add tests and fix bugs for evaluate_gsm8k numeric parser

evaluate_gsm8k.py

@@ -18,15 +18,15 @@ def parse_numeric(s: str):
         return None
 
     # Try fraction first
-    frac_match = re.search(r"(\d+)/(\d+)", s)
+    frac_match = re.search(r"(-?\d+)/(\d+)", s)
     if frac_match:
         try:
             return float(Fraction(int(frac_match.group(1)), int(frac_match.group(2))))
         except Exception:
             pass
 
     # Find decimals or integers
-    nums = re.findall(r"-?\d+\.?\d*", s)
+    nums = re.findall(r"-?\d*\.?\d+", s)
     if not nums:
         return None
 

test_evaluate_gsm8k.py

@@ -0,0 +1,57 @@
+import sys
+from unittest.mock import MagicMock
+
+# Mock out heavy dependencies that might be missing in sandbox
+sys.modules['datasets'] = MagicMock()
+sys.modules['inference'] = MagicMock()
+
+import pytest
+from evaluate_gsm8k import parse_numeric
+
+def test_parse_numeric_valid_integers():
+    """Test parsing of valid positive and negative integers."""
+    assert parse_numeric("42") == 42.0
+    assert parse_numeric("-42") == -42.0
+    assert parse_numeric("0") == 0.0
+
+def test_parse_numeric_valid_decimals():
+    """Test parsing of valid positive and negative decimal numbers."""
+    assert parse_numeric("3.14") == 3.14
+    assert parse_numeric("-0.5") == -0.5
+    assert parse_numeric(".5") == 0.5
+    assert parse_numeric("-.5") == -0.5
+
+def test_parse_numeric_valid_fractions():
+    """Test parsing of fractions, which are expected in GSM8K answers."""
+    assert parse_numeric("1/2") == 0.5
+    assert parse_numeric("-1/2") == -0.5
+    assert parse_numeric("3/4") == 0.75
+
+def test_parse_numeric_invalid_fractions():
+    """Test fractions with zero in the denominator, which should gracefully fall back."""
+    # "1/0" has "1" and "0" as number matches. The last number matched is "0", so it returns 0.0.
+    assert parse_numeric("1/0") == 0.0
+
+def test_parse_numeric_with_text():
+    """Test extracting numbers from strings containing text."""
+    assert parse_numeric("The answer is 42.") == 42.0
+    assert parse_numeric("My fraction is 3/4 and that's it.") == 0.75
+
+def test_parse_numeric_none_or_empty():
+    """Test None, empty, and whitespace strings."""
+    assert parse_numeric(None) is None
+    assert parse_numeric("") is None
+    assert parse_numeric("   ") is None
+    assert parse_numeric("No numbers here!") is None
+
+def test_parse_numeric_multiple_numbers():
+    """Test strings with multiple numbers. Should prefer the last numeric token."""
+    assert parse_numeric("First 10 then 20.") == 20.0
+    # re.search finds the first fraction.
+    assert parse_numeric("I have 1/2 and also 3/4.") == 0.5
+
+def test_parse_numeric_invalid_types():
+    """Test behavior with non-string types."""
+    assert parse_numeric([]) is None
+    assert parse_numeric({}) is None
+    assert parse_numeric(123) is None