diff --git a/task_decomposition/analysis/compare_GPT_groundtruth.py b/task_decomposition/analysis/compare_GPT_groundtruth.py
index 1cb6d86..9ccde75 100644
--- a/task_decomposition/analysis/compare_GPT_groundtruth.py
+++ b/task_decomposition/analysis/compare_GPT_groundtruth.py
@@ -202,6 +202,56 @@ def subtask_similarity(
     score["total"] = TEMPORAL_WEIGHT * temporal_score + SEMANTIC_WEIGHT * semantic_score
     return score
 
+def subtask_similarity2(
+        subtask_decomp_A: list, subtask_decomp_B: list, DEBUG: bool = True
+):
+     """Com"""
+    assert len(subtask_decomp_A) > 0 and len(subtask_decomp_B) > 0
+    assert subtask_decomp_A[0][START_IDX] == subtask_decomp_B[0][START_IDX]
+    assert subtask_decomp_A[-1][END_IDX] == subtask_decomp_B[-1][END_IDX]
+
+    TEMPORAL_WEIGHT = 0.5
+    SEMANTIC_WEIGHT = 0.5
+
+    # TODO: assert subtask_decomp_A and subtask_decomp_B are non-overlapping
+    N = subtask_decomp_A[-1][END_IDX] + 1  # Assuming the last index is end index
+    temporal_score = 0
+    semantic_score = 0
+    temporal_weight_sum = 0
+    for subtask_a in subtask_decomp_A:
+        for subtask_b in subtask_decomp_B:
+            if intersection(subtask_a, subtask_b):
+                IOU = get_IOU(subtask_a, subtask_b)
+                _temporal = IOU
+                # REMOVED IOU
+                _semantic = (
+                    semantic_distance(
+                        subtask_a[SUBTASK_NAME_IDX], subtask_b[SUBTASK_NAME_IDX]
+                    )
+                )
+                # Apply weight based on intersection length relative to total task length
+                temporal_weight = (
+                    max(subtask_a[END_IDX], subtask_b[END_IDX])
+                    - min(subtask_a[START_IDX], subtask_b[START_IDX])
+                    + 1
+                ) / N
+                # semantic_weight = (
+                #     min(subtask_a[END_IDX], subtask_b[END_IDX])
+                #     - max(subtask_a[START_IDX], subtask_b[START_IDX])
+                #     + 1
+                # ) / N
+                temporal_score += _temporal * temporal_weight
+                semantic_score += (
+                    _semantic * temporal_weight
+                )  # or without IOU and just semantic_weight
+                temporal_weight_sum += temporal_weight
+
+    score = {}
+    # Normalized the score with the temporal weight sum
+    score["temporal"] = temporal_score/temporal_weight_sum
+    score["semantic"] = semantic_score/temporal_weight_sum
+    score["total"] = TEMPORAL_WEIGHT * temporal_score + SEMANTIC_WEIGHT * semantic_score
+    return score
 
 # %%
 def test():