Update evaluate.py

evaluate.py

@@ -85,8 +85,19 @@ def load_test_fixtures():
     global test_fixtures
     test_fixtures = []
     env_path = os.environ.get("TEST_FIXTURES_PATH", "").strip()
+
+    # --- START: DEFINITIVE FIX ---
+    # The old code used a relative path, which is unreliable.
+    # This new code builds an absolute path to the fixture file based on
+    # the location of this (evaluate.py) script.
+    script_dir = Path(__file__).parent
+    default_fixture_file = script_dir / "small_test_cases_v10.jsonl"
+
+    candidates = [env_path] if env_path else [str(default_fixture_file)]
+    # --- END: DEFINITIVE FIX ---
     # candidates = [env_path] if env_path else ["conversation_test_fixtures_v10.jsonl"]
-    candidates = [env_path] if env_path else ["small_test_cases_v10.jsonl"]
+    # candidates = [env_path] if env_path else ["small_test_cases_v10.jsonl"]
+    
     path = next((p for p in candidates if p and os.path.exists(p)), None)
     if not path:
         print("Warning: No test fixtures file found for evaluation.")
@@ -105,6 +116,7 @@ def load_test_fixtures():
                 print(f"Skipping malformed JSON line in {path}")
     print(f"Loaded {len(test_fixtures)} fixtures for evaluation from {path}")
 
+    
 def evaluate_nlu_tags(expected: Dict[str, Any], actual: Dict[str, Any], tag_key: str, expected_key_override: str = None) -> Dict[str, float]:
     lookup_key = expected_key_override or tag_key
     expected_raw = expected.get(lookup_key, [])