diff --git a/mtc/ingredient_graph/analyse_graph_scores.py b/mtc/ingredient_graph/analyse_graph_scores.py
index ec938bb..7ff261d 100755
--- a/mtc/ingredient_graph/analyse_graph_scores.py
+++ b/mtc/ingredient_graph/analyse_graph_scores.py
@@ -1,35 +1,41 @@
 from pathlib import Path
 
 import pandas as pd
 import numpy as np
 
 from mtc.settings import NLP_EXPERIMENT_PATH
 
 
+epsilon = 0.001
+
+
 def analyse_graph_scores(time_stamp):
     submission_folder = Path(Path(NLP_EXPERIMENT_PATH) / 'submission_generation' / time_stamp)
     step2_test = pd.read_csv(submission_folder / 'normal' / 'step2_test_scores.csv', header=None).to_numpy().flatten()
     step4_test = pd.read_csv(submission_folder / 'normal' / 'step4_test_scores.csv', header=None).to_numpy().flatten()
 
-    old_scores = step2_test[step2_test != step4_test]
-    new_scores = step4_test[step2_test != step4_test]
+    old_scores = step2_test[abs(step2_test - step4_test) > epsilon]
+    new_scores = step4_test[abs(step2_test - step4_test) > epsilon]
 
+    print('Effect of the score replacement in the test set')
     print(f'n = {len(old_scores)}')
-    print(f'mean scores before replacement: {round(np.mean(old_scores), 2)}')
-    print(f'mean scores after replacement: {round(np.mean(new_scores), 2)}')
+    print(f'before replacement: (mean,std) = {round(np.mean(old_scores), 2), round(np.std(old_scores), 2)}')
+    print(f'after replacement: (mean,std) = {round(np.mean(new_scores), 2), round(np.std(new_scores), 2)}\n')
 
 
 def ratio_graph_scores(time_stamp):
     submission_folder = Path(Path(NLP_EXPERIMENT_PATH) / 'submission_generation' / time_stamp)
 
-    for type in ['train', 'test']:
-        step2 = pd.read_csv(submission_folder / 'normal' / f'step2_{type}_scores.csv', header=None).to_numpy().flatten()
-        step4 = pd.read_csv(submission_folder / 'normal' / f'step4_{type}_scores.csv', header=None).to_numpy().flatten()
+    print('Graph score ratio')
+    for run_type in ['train', 'test']:
+        step2 = pd.read_csv(submission_folder / 'normal' / f'step2_{run_type}_scores.csv', header=None).to_numpy().flatten()
+        step4 = pd.read_csv(submission_folder / 'normal' / f'step4_{run_type}_scores.csv', header=None).to_numpy().flatten()
+
+        scores = step2[abs(step2 - step4) > epsilon]
+        print(f'n_{run_type} = {len(scores)} / {len(step2)} = {len(scores) / len(step2)}')
 
-        scores = step2[step2 != step4]
-        print(f'n_{type} = {len(scores)} / {len(step2)} = {len(scores) / len(step2)}')
 
 if __name__ == '__main__':
-    time_stamp = '03_12_2020_20_18_37'
+    time_stamp = '03_12_2020_20_18_37_original_data'
     analyse_graph_scores(time_stamp)
     ratio_graph_scores(time_stamp)