MrPotato commited on
Commit
0992e03
·
1 Parent(s): 989689c

changed tokenization

Browse files
Files changed (1) hide show
  1. ref_seg_ger.py +8 -3
ref_seg_ger.py CHANGED
@@ -19,7 +19,7 @@ import os
19
  import numpy as np
20
  from PIL import Image
21
  from tokenizers import pre_tokenizers
22
- from tokenizers.pre_tokenizers import Digits, Split, Whitespace
23
  import datasets
24
  from itertools import chain
25
  import pandas as pd
@@ -143,9 +143,14 @@ class RefSeg(datasets.GeneratorBasedBuilder):
143
 
144
  # DEFAULT_CONFIG_NAME = "small" # It's not mandatory to have a default configuration. Just use one if it make sense.
145
 
146
- #split_tokens = [".", ":", ",",";","/","-","(", ")"]
147
 
148
- TOKENIZER = Whitespace()
 
 
 
 
 
149
 
150
  def _info(self):
151
  # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
 
19
  import numpy as np
20
  from PIL import Image
21
  from tokenizers import pre_tokenizers
22
+ from tokenizers.pre_tokenizers import Digits, Split, Whitespace, Sequence
23
  import datasets
24
  from itertools import chain
25
  import pandas as pd
 
143
 
144
  # DEFAULT_CONFIG_NAME = "small" # It's not mandatory to have a default configuration. Just use one if it make sense.
145
 
146
+ split_tokens = [".", ":", ",", ";", "/", "-", "(", ")"]
147
 
148
+ TOKENIZER = Sequence([
149
+ Whitespace(),
150
+ Digits(),
151
+ ] + [Split(x, behavior="isolated") for x in split_tokens])
152
+
153
+ #TOKENIZER = Whitespace()
154
 
155
  def _info(self):
156
  # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset