singh-aditya
commited on
Commit
•
004ae91
1
Parent(s):
ffa5c9b
Create create_dataset.py
Browse filesFile to convert the dataset into JSON format so that it can be easily loaded in the Huggingface dataset.
- create_dataset.py +162 -0
create_dataset.py
ADDED
@@ -0,0 +1,162 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import json
|
3 |
+
import glob
|
4 |
+
from tqdm import tqdm
|
5 |
+
from io import StringIO
|
6 |
+
import pandas as pd
|
7 |
+
|
8 |
+
"""
|
9 |
+
Steps to convert the data into JSON format.
|
10 |
+
|
11 |
+
Step-0: Use a python environment where pandas is installed.
|
12 |
+
Step-1: Download the source file from here: https://figshare.com/articles/dataset/MACCROBAT2018/9764942
|
13 |
+
Step-2: Unzip the file in put that into a folder (say `data` folder).
|
14 |
+
All unzipped files will be present here.
|
15 |
+
* data/MACCROBAT2020/*.txt
|
16 |
+
* data/MACCROBAT2020/*.ann
|
17 |
+
Step-3: Use the correct paths and run this file.
|
18 |
+
"""
|
19 |
+
|
20 |
+
def remove_overlapped_ner_tags(ner_details: list[dict]):
|
21 |
+
"""remove overlapping entities.
|
22 |
+
|
23 |
+
Args:
|
24 |
+
ner_details (List[dict]): a list of dictionary where each dictionary holds
|
25 |
+
the information of a entity.
|
26 |
+
|
27 |
+
NOTE: Priority is given to the entity that is labelled first after sorting all by start index in ascending order.
|
28 |
+
(i.e. it's end-index is less than other start of other overlapping entity.)
|
29 |
+
|
30 |
+
Returns:
|
31 |
+
list[dict]: updated list (removed item if something was overlapping)
|
32 |
+
"""
|
33 |
+
# funtion to remove the overlapping NER-tags
|
34 |
+
new_ner_details = []
|
35 |
+
|
36 |
+
ner_details = sorted(ner_details, key=lambda x: x["start"])
|
37 |
+
for i, ner_detail in enumerate(ner_details):
|
38 |
+
if i == 0:
|
39 |
+
start = ner_detail["start"]
|
40 |
+
end = ner_detail["end"]
|
41 |
+
new_ner_details.append(ner_detail)
|
42 |
+
continue
|
43 |
+
|
44 |
+
current_start = ner_detail["start"]
|
45 |
+
current_end = ner_detail["end"]
|
46 |
+
if current_start < end:
|
47 |
+
continue
|
48 |
+
|
49 |
+
# update the start and end
|
50 |
+
start = current_start
|
51 |
+
end = current_end
|
52 |
+
|
53 |
+
new_ner_details.append(ner_detail)
|
54 |
+
return new_ner_details
|
55 |
+
|
56 |
+
|
57 |
+
def get_ner_details(ann_file):
|
58 |
+
with open(ann_file, "r") as f:
|
59 |
+
lines = f.readlines()
|
60 |
+
lines = [line.strip() for line in lines]
|
61 |
+
csv_data = "\n".join(lines)
|
62 |
+
csv_data = StringIO(csv_data)
|
63 |
+
df = pd.read_csv(csv_data, sep="\t", header=None)
|
64 |
+
|
65 |
+
df.columns = ["EntityID", "EntityDetails", "EntityText"]
|
66 |
+
# print(df.shape)
|
67 |
+
|
68 |
+
# remove rows where entity-id start other than `T`
|
69 |
+
df = df[df["EntityID"].apply(lambda x: str(x).strip().startswith("T"))]
|
70 |
+
|
71 |
+
# remove the rows which contains the ";" in the `EntityDetails`
|
72 |
+
df = df[df["EntityDetails"].apply(lambda x: ";" not in str(x))]
|
73 |
+
|
74 |
+
# drop where None is present
|
75 |
+
df.dropna(axis=1, inplace=True)
|
76 |
+
|
77 |
+
ner_info = []
|
78 |
+
for i, row in df.iterrows():
|
79 |
+
text = row["EntityText"]
|
80 |
+
details = row["EntityDetails"]
|
81 |
+
try:
|
82 |
+
ner_tag, start, end = details.split(" ")
|
83 |
+
except:
|
84 |
+
print(ann_file)
|
85 |
+
print(details)
|
86 |
+
start = int(float(start))
|
87 |
+
end = int(float(end))
|
88 |
+
|
89 |
+
ner_info.append({"text": text, "label": ner_tag.upper(), "start": start, "end": end})
|
90 |
+
|
91 |
+
# remove the overlapping entities
|
92 |
+
ner_info = remove_overlapped_ner_tags(ner_details=ner_info)
|
93 |
+
# print(ner_info)
|
94 |
+
|
95 |
+
return ner_info
|
96 |
+
|
97 |
+
|
98 |
+
def main(input_path: str = "data/MACCROBAT2020", output_path: str = "data/MACCROBAT2020-V2.json"):
|
99 |
+
txt_files = glob.glob(os.path.join(input_path, "*.txt"))
|
100 |
+
txt_files.sort()
|
101 |
+
|
102 |
+
ner_data = {"data": [], "verson": "MACCROBAT-V2 (https://figshare.com/articles/dataset/MACCROBAT2018/9764942)"}
|
103 |
+
|
104 |
+
for txt_file in tqdm(txt_files, desc="Extracting data..."):
|
105 |
+
with open(txt_file, "r") as f:
|
106 |
+
full_text = f.read()
|
107 |
+
a = txt_file.replace(".txt", ".ann")
|
108 |
+
ner_info = get_ner_details(a)
|
109 |
+
data = {"full_text": full_text, "ner_info": ner_info}
|
110 |
+
ner_data["data"].append(data)
|
111 |
+
|
112 |
+
ALL_NER_LABLES = set()
|
113 |
+
for details in tqdm(ner_data["data"], desc="Splitting into tokens..."):
|
114 |
+
text = details["full_text"]
|
115 |
+
ner_details = details["ner_info"]
|
116 |
+
|
117 |
+
tokens = []
|
118 |
+
ner_labels = []
|
119 |
+
start = 0
|
120 |
+
for ner_detail in ner_details:
|
121 |
+
ner_start = ner_detail["start"]
|
122 |
+
ner_end = ner_detail["end"]
|
123 |
+
|
124 |
+
before_ner_token = text[start:ner_start]
|
125 |
+
ner_token = text[ner_start:ner_end]
|
126 |
+
|
127 |
+
tokens.append(before_ner_token)
|
128 |
+
ner_labels.append("O")
|
129 |
+
tokens.append(ner_token)
|
130 |
+
ner_labels.append(f'B-{ner_detail["label"]}')
|
131 |
+
ALL_NER_LABLES.add(f'B-{ner_detail["label"]}')
|
132 |
+
ALL_NER_LABLES.add(f'I-{ner_detail["label"]}')
|
133 |
+
start = ner_end
|
134 |
+
if len(text) >= start:
|
135 |
+
ner_labels.append("O")
|
136 |
+
tokens.append(text[start:])
|
137 |
+
assert len(tokens) == len(ner_labels)
|
138 |
+
|
139 |
+
details["tokens"] = tokens
|
140 |
+
details["ner_labels"] = ner_labels
|
141 |
+
|
142 |
+
ner_data["all_ner_labels"] = sorted(list(ALL_NER_LABLES), key=lambda x: x.split("-")[-1])
|
143 |
+
|
144 |
+
label_2_index = {k: i for i, k in enumerate(ner_data["all_ner_labels"])}
|
145 |
+
index_2_label = {v: k for k, v in label_2_index.items()}
|
146 |
+
ner_data["label_2_index"] = label_2_index
|
147 |
+
ner_data["index_2_label"] = index_2_label
|
148 |
+
|
149 |
+
for details in tqdm(ner_data["data"], desc="label2index..."):
|
150 |
+
ner_labels = details["ner_labels"]
|
151 |
+
ner_labels_ids = []
|
152 |
+
for ner in ner_labels:
|
153 |
+
ner_labels_ids.append(label_2_index.get(ner))
|
154 |
+
details["ner_labels"] = ner_labels_ids
|
155 |
+
with open(output_path, "w") as f:
|
156 |
+
json.dump(ner_data, f, indent=4)
|
157 |
+
|
158 |
+
|
159 |
+
if __name__ == "__main__":
|
160 |
+
input_path: str = "data/MACCROBAT2020"
|
161 |
+
output_path: str = "data/MACCROBAT2020-V2.json"
|
162 |
+
main(input_path=input_path, output_path=output_path)
|