Refactored test cases and added new metrics

tests.py

@@ -1,17 +1,36 @@
+# test_cases = [
+#     {
+#         "predictions": [0, 0],
+#         "references": [1, 1],
+#         "result": {"metric_score": 0}
+#     },
+#     {
+#         "predictions": [1, 1],
+#         "references": [1, 1],
+#         "result": {"metric_score": 1}
+#     },
+#     {
+#         "predictions": [1, 0],
+#         "references": [1, 1],
+#         "result": {"metric_score": 0.5}
+#     }
+# ]
+
 test_cases = [
     {
-        "predictions": [0, 0],
-        "references": [1, 1],
-        "result": {"metric_score": 0}
+        "predictions": ["1111", "1113", "1120", "1211"],
+        "references": ["1111", "1112", "1113", "1114"],
+        "result": {"metric_score": 0},
     },
     {
         "predictions": [1, 1],
         "references": [1, 1],
-        "result": {"metric_score": 1}
+        "result": {
+            "accuracy": 0.25,
+            "hierarchical_precision": 0.7142857142857143,
+            "hierarchical_recall": 0.5,
+            "hierarchical_fmeasure": 0.588235294117647,
+        },
     },
-    {
-        "predictions": [1, 0],
-        "references": [1, 1],
-        "result": {"metric_score": 0.5}
-    }
-]
+    # {"predictions": [1, 0], "references": [1, 1], "result": {"metric_score": 0.5}},
+]