removed dependencies (bigbench library, tensorflow, protobuf)

Author: sileod

src/tasksource/access.py

@@ -6,6 +6,7 @@
 from datasets import load_dataset
 import funcy as fc
 import os
+import copy
 from sorcery import dict_of
 
 def parse_var_name(s):
@@ -63,8 +64,8 @@ def dict_to_query(d=dict(), **kwargs):
     return '&'.join([f'`{k}`=="{v}"' for k,v in d.items()])
 
 def load_preprocessing(tasks=tasks, **kwargs):
-    y = task_df.query(dict_to_query(**kwargs)).iloc[0]
-    preprocessing= getattr(tasks, y.preprocessing_name)
+    y = task_df.copy().query(dict_to_query(**kwargs)).iloc[0]
+    preprocessing= copy.deepcopy(getattr(tasks, y.preprocessing_name))
     for c in 'dataset_name','config_name':
         if not isinstance(getattr(preprocessing,c), str):
              setattr(preprocessing,c,getattr(y,c))

src/tasksource/tasks.py

@@ -181,6 +181,7 @@ def _imppres_post_process(ds,prefix=''):
     'inputs',
     choices_list='multiple_choice_targets',
     labels=lambda x:x['multiple_choice_scores'].index(1) if 1 in ['multiple_choice_scores'] else -1,
+    dataset_name='tasksource/bigbench',
     config_name=bigbench_discriminative_english - {"social_i_qa","intersect_geometry"} # english multiple choice tasks, minus duplicates
 )
 
@@ -715,4 +716,8 @@ def _preprocess_chatgpt_detection(ex):
 
 
 #autotnli = Classification("premises", "hypothesis", "label", dataset_name="metaeval/autotnli")
-#equate = Classification("sentence1", "sentence2", "gold_label",dataset_name="metaeval/equate")
+#equate = Classification("sentence1", "sentence2", "gold_label",dataset_name="metaeval/equate")
+
+conqada = Classification("sentence1","sentence2","label",dataset_name="lasha-nlp/CONDAQA",
+    pre_process = lambda ds:ds.filter(lambda x:x['label'] in {"DON'T KNOW","YES","NO"})
+)