Update vasr.py

vasr.py

@@ -101,9 +101,7 @@ class Winogavil(datasets.GeneratorBasedBuilder):
 
         df = pd.read_csv(examples_csv)
 
-        # columns_to_serialize = ['candidates', 'associations']
-        # for c in columns_to_serialize:
-        #   df[c] = df[c].apply(json.loads)
+        d_keys = ['A_img', 'B_img', 'C_img', 'candidates', 'label', 'D_img', 'A_verb', 'B_verb', 'C_verb', 'D_verb', 'diff_item_A', 'diff_item_A_str_first', 'diff_item_B', 'diff_item_B_str_first']
 
         for r_idx, r in df.iterrows():
             r_dict = r.to_dict()
@@ -111,4 +109,5 @@ class Winogavil(datasets.GeneratorBasedBuilder):
             candidates_images = [os.path.join(images_dir, "vasr_images", f"{x}.{self.IMAGE_EXTENSION}") for x in
                                  r_dict['candidates']]
             r_dict['candidates_images'] = candidates_images
-            yield r_idx, r_dict
+            relevant_r_dict = {k:v for k,v in r_dict.items() if k in d_keys or k == 'candidates_images'}
+            yield r_idx, relevant_r_dict