michal-stefanik
commited on
Commit
•
e8fa15d
1
Parent(s):
ac31af3
Upload parse_czech_squad.py
Browse files- parse_czech_squad.py +109 -0
parse_czech_squad.py
ADDED
@@ -0,0 +1,109 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import csv
|
2 |
+
import json
|
3 |
+
import os.path
|
4 |
+
from typing import Iterable
|
5 |
+
|
6 |
+
import pandas as pd
|
7 |
+
|
8 |
+
data_folder = "data/czech-squad-v3"
|
9 |
+
|
10 |
+
shorten_to_sentences = 3
|
11 |
+
|
12 |
+
out_json = "data/czech_squad_%s-sents.json" % shorten_to_sentences
|
13 |
+
|
14 |
+
|
15 |
+
def read_first_entries(fpath: str, sep: str = "\t"):
|
16 |
+
line_collector = []
|
17 |
+
|
18 |
+
with open(fpath) as f:
|
19 |
+
for line in f.readlines():
|
20 |
+
entry = line.split(sep)[0]
|
21 |
+
line_collector.append(entry)
|
22 |
+
|
23 |
+
return line_collector
|
24 |
+
|
25 |
+
|
26 |
+
def collect_tokens(s: Iterable[str]) -> str:
|
27 |
+
out_str = ""
|
28 |
+
last_g = False
|
29 |
+
for i, token in enumerate(s):
|
30 |
+
token = token.strip()
|
31 |
+
if token is None:
|
32 |
+
raise ValueError("Token on position %s is None" % i)
|
33 |
+
if token == "<g/>":
|
34 |
+
last_g = True
|
35 |
+
continue
|
36 |
+
elif token.startswith("<") and token.endswith(">"):
|
37 |
+
continue
|
38 |
+
else:
|
39 |
+
if last_g:
|
40 |
+
out_str += token
|
41 |
+
last_g = False
|
42 |
+
else:
|
43 |
+
out_str += " %s" % token
|
44 |
+
return out_str.strip()
|
45 |
+
|
46 |
+
|
47 |
+
out_dict = {}
|
48 |
+
|
49 |
+
for i, folder in enumerate(os.listdir(data_folder)):
|
50 |
+
try:
|
51 |
+
question_f = os.path.join(data_folder, folder, "01question.vert")
|
52 |
+
question_list = read_first_entries(question_f)
|
53 |
+
question_str = collect_tokens(question_list)
|
54 |
+
|
55 |
+
# reformulated answer selection
|
56 |
+
# answer_f = os.path.join(data_folder, folder, "02answer.vert")
|
57 |
+
# answer_list = read_first_entries(answer_f)
|
58 |
+
# # answer_df = pd.read_csv(answer_f, sep="\t", index_col=False)
|
59 |
+
# answer_str = collect_tokens(answer_list)
|
60 |
+
|
61 |
+
answer_f = os.path.join(data_folder, folder, "09answer_extraction.vert")
|
62 |
+
answer_list = read_first_entries(answer_f)
|
63 |
+
# answer_df = pd.read_csv(answer_f, sep="\t", index_col=False)
|
64 |
+
answer_str = collect_tokens(answer_list)
|
65 |
+
answer_str = answer_str.split(" # ")[0]
|
66 |
+
|
67 |
+
answer_type_f = os.path.join(data_folder, folder, "05metadata.txt")
|
68 |
+
answer_type = next(t for t in read_first_entries(answer_type_f) if "a_type" in t)
|
69 |
+
answer_type_cleaned = answer_type.replace("<a_type>", "").replace("</a_type>", "").strip()
|
70 |
+
|
71 |
+
text_f = os.path.join(data_folder, folder, "03text.vert")
|
72 |
+
text_list = read_first_entries(text_f)
|
73 |
+
# text_df = pd.read_csv(text_f, sep="\t", engine="python", error_bad_lines=False)
|
74 |
+
text_str = collect_tokens(text_list)
|
75 |
+
|
76 |
+
if answer_str.lower() not in text_str.lower():
|
77 |
+
print("Skipping answer %s: not present in context." % answer_str)
|
78 |
+
continue
|
79 |
+
|
80 |
+
if answer_str.endswith("."):
|
81 |
+
# to match in multi-sentence matching
|
82 |
+
answer_str = answer_str[:-1]
|
83 |
+
|
84 |
+
# maybe shorten to n-surrounding sentences
|
85 |
+
if shorten_to_sentences is not None:
|
86 |
+
sentences = text_str.split(". ")
|
87 |
+
answer_sentence_idx = next(i for i, _ in enumerate(sentences)
|
88 |
+
if all(a_segment.lower() in sentences[i+j].lower()
|
89 |
+
for j, a_segment in enumerate(answer_str.split(". "))))
|
90 |
+
shortened_context = sentences[max(0, answer_sentence_idx - shorten_to_sentences):
|
91 |
+
min(len(sentences), answer_sentence_idx + shorten_to_sentences)]
|
92 |
+
|
93 |
+
text_str = ". ".join(shortened_context) + ". "
|
94 |
+
|
95 |
+
# TODO: squad-like format: https://huggingface.co/datasets/squad
|
96 |
+
out_dict[i] = {"id": folder.split("/")[-1],
|
97 |
+
"answer_type": answer_type_cleaned,
|
98 |
+
"context": text_str,
|
99 |
+
"question": question_str,
|
100 |
+
"answers": {"text": [answer_str]}
|
101 |
+
}
|
102 |
+
|
103 |
+
except NotADirectoryError as e:
|
104 |
+
print("Skipping %s: %s: %s" % (i, folder, e))
|
105 |
+
|
106 |
+
with open(out_json, "w") as out_f:
|
107 |
+
out_f.write(json.dumps(out_dict))
|
108 |
+
|
109 |
+
print("Done. Output json exported to %s" % out_json)
|