Datasets:

ArXiv:
Tags:
License:
Muennighoff commited on
Commit
9dd5ea5
1 Parent(s): 6aa2c6d

Add cluewsc2020 merging script

Browse files
Files changed (1) hide show
  1. raw_data/merge_clue_cluewsc2020.py +136 -0
raw_data/merge_clue_cluewsc2020.py ADDED
@@ -0,0 +1,136 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ """
2
+ Adds clue/cluewsc2020 samples to xwinograd
3
+ From: https://gist.github.com/jordiclive/26506ea7e897ad8270f9e793bdc285b5
4
+ """
5
+
6
+ import json
7
+
8
+ import datasets
9
+ import pandas as pd
10
+ from datasets import load_dataset
11
+
12
+
13
+ def find_pronoun(x):
14
+ pronoun = x["target"]["span2_text"]
15
+ indices = [x["target"]["span2_index"], x["target"]["span2_index"] + len(pronoun)]
16
+ return [pronoun, indices, list(pronoun)]
17
+
18
+
19
+ def find_switch(x):
20
+ pronoun = x["target"]["span1_text"]
21
+ indices = [x["target"]["span1_index"], x["target"]["span1_index"] + len(pronoun)]
22
+ if x["label"] == 1:
23
+ label = False
24
+ else:
25
+ label = True
26
+ return [pronoun, indices, list(pronoun), label]
27
+
28
+
29
+ def convert_to_format(df):
30
+
31
+ df["pronoun"] = df.apply(find_pronoun, axis=1)
32
+
33
+ df["toks"] = df["text"].apply(lambda x: list(x))
34
+
35
+ df["switch"] = df.apply(find_switch, axis=1)
36
+ df.reset_index(inplace=True, drop=True)
37
+
38
+ lang = []
39
+ original = []
40
+ o_text = []
41
+ sent = []
42
+ toks = []
43
+ pronoun = []
44
+ switch = []
45
+ df["pronoun_to_replace"] = df["target"].apply(lambda x: x["span2_text"])
46
+ for i, df_text in df.groupby(["text", "pronoun_to_replace"]):
47
+ if len(df_text) == 1:
48
+ continue
49
+ df_text.reset_index(inplace=True, drop=True)
50
+ try:
51
+ if df_text["label"][0] != df_text["label"][1]:
52
+ df_text = df_text[:2]
53
+ elif df_text["label"][0] != df_text["label"][2] and len(df_text) > 2:
54
+ df_text = df_text.iloc[[0, 2], :]
55
+ df_text.reset_index(inplace=True, drop=True)
56
+ df_new = df_text[:1]
57
+ df_new["switch"] = df_new["switch"].apply(
58
+ lambda x: [df_text["switch"][0], df_text["switch"][1]]
59
+ )
60
+
61
+ lang.append("zh")
62
+ original.append("original")
63
+ o_text.append("?")
64
+ sent.append(df_new.iloc[0]["text"])
65
+ toks.append(df_new.iloc[0]["toks"])
66
+ pronoun.append(df_new.iloc[0]["pronoun"])
67
+ switch.append(df_new.iloc[0]["switch"])
68
+
69
+ except:
70
+ continue
71
+
72
+ total_df = pd.DataFrame(
73
+ {0: lang, 1: original, 2: o_text, 3: sent, 4: toks, 5: pronoun, 6: switch}
74
+ )
75
+ count = total_df[5].apply(lambda x: len(x[0]))
76
+ total_df[4] = total_df[4].apply(lambda x: json.dumps(x))
77
+ total_df[5] = total_df[5].apply(lambda x: json.dumps(x))
78
+ total_df[6] = total_df[6].apply(lambda x: json.dumps(x))
79
+ return total_df, count
80
+
81
+
82
+ def remove_at(i, s):
83
+ return s[:i] + s[i + 1 :]
84
+
85
+
86
+ def remove_to(x):
87
+ if x["count"] == 2:
88
+ return remove_at(x["sentence"].index("_") + 1, x["sentence"])
89
+ else:
90
+ return x["sentence"]
91
+
92
+
93
+ def get_original_splits():
94
+ # xwinograd directory is https://huggingface.co/datasets/Muennighoff/xwinograd/tree/9dbcc59f86f9e53e0b36480d806982499d877edc
95
+
96
+ for j, i in enumerate(["en", "jp", "ru", "pt", "fr", "zh"]):
97
+ if j == 0:
98
+ dfx = datasets.load_dataset("xwinograd", i)["test"].to_pandas()
99
+ dfx["lang"] = i
100
+ else:
101
+ df = datasets.load_dataset("xwinograd", i)["test"].to_pandas()
102
+ df["lang"] = i
103
+ dfx = pd.concat([dfx, df])
104
+ return dfx
105
+
106
+
107
+ def get_examples_from_clue():
108
+ # xwinograd directory is commit: https://huggingface.co/datasets/Muennighoff/xwinograd/tree/9dbcc59f86f9e53e0b36480d806982499d877edc
109
+
110
+ dataset = load_dataset("clue", "cluewsc2020")
111
+ df = dataset["train"].to_pandas()
112
+ df_val = dataset["validation"].to_pandas()
113
+ df = pd.concat([df, df_val])
114
+ new_examples, count = convert_to_format(df)
115
+ new_examples.reset_index(inplace=True, drop=True)
116
+ new_examples.to_csv(
117
+ "xwinograd/data/xwinograd.tsv", sep="\t", header=None, index=False
118
+ )
119
+ df_post = datasets.load_dataset("xwinograd", "zh")
120
+ df_post = df_post["test"].to_pandas()
121
+ df_post["count"] = count
122
+ df_post["sentence"] = df_post.apply(remove_to, axis=1)
123
+ df_post = df_post[["sentence", "option1", "option2", "answer"]]
124
+ return df_post
125
+
126
+
127
+ if __name__ == "__main__":
128
+ # xwinograd directory is commit: https://huggingface.co/datasets/Muennighoff/xwinograd/tree/9dbcc59f86f9e53e0b36480d806982499d877edc
129
+ dfx = get_original_splits()
130
+ df_post = get_examples_from_clue()
131
+ df_post.to_json("new_examples_updated.json", orient="split")
132
+ df_post["lang"] = "zh"
133
+ dfx = pd.concat([dfx, df_post])
134
+ dfx.drop_duplicates()
135
+ dfx.reset_index(inplace=True, drop=True)
136
+ dfx.to_json("all_examples_updated.json", orient="split")