Sigurdur commited on
Commit
365d2bf
1 Parent(s): 027c6ba

Upload make_hf_dataset.py

Browse files
Files changed (1) hide show
  1. make_hf_dataset.py +65 -0
make_hf_dataset.py ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from datasets import Dataset, DatasetDict, load_dataset, concatenate_datasets
2
+ import pandas as pd
3
+ import os
4
+
5
+
6
+ def prepare_nqii(path, split="train"):
7
+ ds = load_dataset(path)
8
+ ds = ds[split]
9
+
10
+ # to dataframe
11
+ df = pd.DataFrame(ds)
12
+ df = df.drop(columns=["title"])
13
+
14
+ df["start"] = df["answers"].apply(lambda x: x["answer_start"])
15
+ df["answers"] = df["answers"].apply(lambda x: x["text"])
16
+
17
+ # change type
18
+ ds = Dataset.from_pandas(df)
19
+
20
+ return ds
21
+
22
+
23
+ def prepare_ruquad(path):
24
+ df = pd.read_json(path)
25
+
26
+ # filter
27
+ df = df[df["type"] == "ANSWERED_WITH_SPAN"]
28
+
29
+ # drop index
30
+ df = df.reset_index(drop=True)
31
+
32
+ df = df.rename(columns={"question_id":"id", "paragraph": "context", "span":"answers"})
33
+ df = df.drop(columns=[ "type", "answer_id", "article_id", "end", "source", "answer"])
34
+
35
+ df["start"] = df["start"].astype(int)
36
+
37
+ df["answers"] = df["answers"].apply(lambda x: [x])
38
+ df["start"] = df["start"].apply(lambda x: [x])
39
+
40
+ ds = Dataset.from_pandas(df)
41
+
42
+ return ds
43
+
44
+
45
+ def download_ruquad():
46
+ os.system("curl --remote-name-all https://repository.clarin.is/repository/xmlui/bitstream/handle/20.500.12537/310{/RUQuAD1.zip}")
47
+ os.system("unzip RUQuAD1.zip")
48
+
49
+
50
+
51
+ if __name__ == "__main__":
52
+ nqii_train = prepare_nqii("vesteinn/icelandic-qa-NQiI", split="train")
53
+ ruquad_train = prepare_ruquad("train.json")
54
+ train = concatenate_datasets([nqii_train, ruquad_train])
55
+
56
+ val = prepare_nqii("vesteinn/icelandic-qa-NQiI", split="validation")
57
+
58
+ nqi_test = prepare_nqii("vesteinn/icelandic-qa-NQiI", split="test")
59
+ ruquad_test = prepare_ruquad("test.json")
60
+ test = concatenate_datasets([nqi_test, ruquad_test])
61
+
62
+ dataset = DatasetDict({"train": train, "validation": val, "test": test})
63
+ print(dataset)
64
+
65
+ dataset.push_to_hub("Sigurdur/nqii_ruqad")