Datasets:
Upload dataset-preprocessor.py
Browse files- dataset-preprocessor.py +54 -0
dataset-preprocessor.py
ADDED
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from datasets import load_dataset
|
2 |
+
import re
|
3 |
+
import random
|
4 |
+
|
5 |
+
def split_into_paragraphs(text):
|
6 |
+
# Split by markdown headers or double newlines
|
7 |
+
paragraphs = re.split(r'\n\n|(?=^#)', text, flags=re.MULTILINE)
|
8 |
+
return [p.strip() for p in paragraphs if p.strip()]
|
9 |
+
|
10 |
+
def create_input_output_pairs(example):
|
11 |
+
paragraphs = example['paragraphs']
|
12 |
+
n_paragraphs = len(paragraphs)
|
13 |
+
|
14 |
+
# Randomly select about half of the paragraphs for input
|
15 |
+
n_input = max(1, random.randint(n_paragraphs // 2 - 1, n_paragraphs // 2 + 1))
|
16 |
+
|
17 |
+
input_paragraphs = paragraphs[:n_input]
|
18 |
+
output_paragraphs = paragraphs[n_input:]
|
19 |
+
|
20 |
+
return {
|
21 |
+
'inputs': ' '.join(input_paragraphs),
|
22 |
+
'targets': ' '.join(output_paragraphs)
|
23 |
+
}
|
24 |
+
|
25 |
+
def preprocess_dataset(dataset_name, text_column='text'):
|
26 |
+
# Load the dataset
|
27 |
+
dataset = load_dataset(dataset_name)
|
28 |
+
|
29 |
+
# Split text into paragraphs
|
30 |
+
dataset = dataset.map(
|
31 |
+
lambda example: {'paragraphs': split_into_paragraphs(example[text_column])},
|
32 |
+
remove_columns=[text_column]
|
33 |
+
)
|
34 |
+
|
35 |
+
# Create input-output pairs
|
36 |
+
preprocessed_dataset = dataset.map(
|
37 |
+
create_input_output_pairs,
|
38 |
+
remove_columns=['paragraphs']
|
39 |
+
)
|
40 |
+
|
41 |
+
return preprocessed_dataset
|
42 |
+
|
43 |
+
# Usage example
|
44 |
+
if __name__ == "__main__":
|
45 |
+
# Replace 'your_dataset' with the actual dataset name
|
46 |
+
dataset_name = 'your_dataset'
|
47 |
+
|
48 |
+
preprocessed_dataset = preprocess_dataset(dataset_name)
|
49 |
+
|
50 |
+
# Print some examples
|
51 |
+
print(preprocessed_dataset['train'][:5])
|
52 |
+
|
53 |
+
# Save the preprocessed dataset
|
54 |
+
preprocessed_dataset.save_to_disk("preprocessed_dataset")
|