forked from MurtyShikhar/robustqa
-
Notifications
You must be signed in to change notification settings - Fork 0
/
subset.py
59 lines (45 loc) · 1.9 KB
/
subset.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
import args
import json
from collections import Counter
def read_data(directory, datasets):
dataset_dict = {}
for dataset in datasets:
curr_dataset = json.loads(open(f'{directory}/{dataset}').read())
dataset_dict[dataset] = curr_dataset
topics_list = []
for dataset in datasets:
for entry in dataset_dict[dataset]["data"]:
if "topic" in entry:
topics_list.append(entry["topic"])
else:
topics_list.append(dataset)
return dataset_dict, topics_list
def create_subsets(directory, datasets, keep_percentage=1):
print(f'Creating subsets in {directory}')
dataset_dict, topics_list = read_data(directory, datasets)
# compute count of entries per topic
# multiply by subset keep percentage to calculate num of entries to keep
topics_count = dict(Counter(topics_list))
for key in topics_count.keys():
# keep at least 3 per topic
topics_count[key] = max(3, int(topics_count[key] * keep_percentage))
for dataset in datasets:
dataset_subset = []
print(f'{dataset} original size: {len(dataset_dict[dataset]["data"])}')
for entry in dataset_dict[dataset]["data"]:
topic = dataset
if "topic" in entry:
topic = entry["topic"]
if topics_count[topic] > 0:
topics_count[topic] -= 1
dataset_subset.append(entry)
print(f'{dataset} subset size: {len(dataset_subset)}')
subset_dict = {"data": dataset_subset}
with open(f'{directory}/{dataset}_subset', 'w') as f:
json.dump(subset_dict, f)
print()
if __name__ == "__main__":
args = args.get_train_test_args()
datasets = args["train_datasets"].split(',')
create_subsets(args["train_dir"], datasets, args["subset_keep_percentage"])
create_subsets(args["val_dir"], datasets)