Make processing of punctuation consistent with train data: frame each symbol with spaces

model_wrapper.py

@@ -4,6 +4,7 @@ import tempfile
 import sys
 import datetime
 import re
+import string
 sys.path.append('mtool')
 
 import torch
@@ -78,7 +79,10 @@ class PredictionModel:
 
 
     def clean_texts(self, texts):
-        return [re.sub(r' +', ' ', t) for t in texts]
+        punctuation = ''.join([f'\\{s}' for s in string.punctuation])
+        texts = [re.sub(f'([{punctuation}])', ' \\1 ', t) for t in texts]
+        texts = [re.sub(r' +', ' ', t) for t in texts]
+        return texts
 
 
     def _predict_to_mrp(self, texts, graph_mode='labeled-edge'):