ignore single whitespace at beginning and end of answer, use reward = len(oracle_answer) / len(answer)

reasoning_gym/dataset.py

@@ -59,7 +59,7 @@ class ProceduralDataset(ABC, Sized, Iterable[Dict[str, Any]]):
             if answer == oracle_answer:
                 reward = 1.0
             elif oracle_answer in answer:
-                reward = 0.5
+                reward = len(oracle_answer) / len(answer)
             else:
                 reward = 0.01
 

reasoning_gym/utils.py

@@ -17,7 +17,7 @@ Once you have thought about the reasoning process, provide the answer in the fol
 
 
 def extract_answer(completion: str, tag_name: str = "answer") -> Optional[str]:
-    regex = f"<{tag_name}>(.*?)</{tag_name}>"
+    regex = f"<{tag_name}>\\s?(.*?)\\s?</{tag_name}>"
     matches = list(
         re.finditer(
             regex,