diff --git a/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/scripts/macro_avg.py b/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/scripts/macro_avg.py
index 768a1678aa812c1c96affd5c467253d01bee998a..dc0f9ff3a11a8e9572c6daba2415162a9a54a392 100644
--- a/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/scripts/macro_avg.py
+++ b/PaddleNLP/Research/MRQA2019-D-NET/multi_task_learning/scripts/macro_avg.py
@@ -1,13 +1,13 @@
 import numpy as np
 import argparse
+import json
 import re
 
 def extract_score(line):
-    prog = re.compile(r'{"f1": (-?\d+\.?\d*e?-?\d*?), "exact_match": (-?\d+\.?\d*e?-?\d*?)}')
-    result = prog.match(line)
-    f1 = float(result.group(1))
-    em = float(result.group(2))
-    return f1, em
+    score_json = json.loads(line)
+    f1 = score_json['f1']
+    em = score_json['exact_match']
+    return float(f1), float(em)
 
 
 if __name__ == '__main__':
@@ -16,7 +16,7 @@ if __name__ == '__main__':
     parser.add_argument('input_file', help='Score file')
     args = parser.parse_args()
     with open(args.input_file) as fin:
-        lines = map(str.strip, fin.readlines())
+        lines = list(map(str.strip, fin.readlines()))
     in_domain_scores = {}
     for dataset_id in range(0, 12, 2):
         f1, em = extract_score(lines[dataset_id+1])
@@ -25,5 +25,5 @@ if __name__ == '__main__':
     for dataset_id in range(12, 24, 2):
         f1, em = extract_score(lines[dataset_id+1])
         out_of_domain_scores[lines[dataset_id]] = f1
-    print('In domain avg: {}'.format(np.mean(in_domain_scores.values())))
-    print('Out of domain avg: {}'.format(np.mean(out_of_domain_scores.values())))
+    print('In domain avg: {}'.format(sum(in_domain_scores.values()) / len(in_domain_scores.values())))
+    print('Out of domain avg: {}'.format(sum(out_of_domain_scores.values()) / len(in_domain_scores.values())))