Upload generate_triplets.py
Browse files- generate_triplets.py +68 -0
generate_triplets.py
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import itertools
|
3 |
+
from datasets import load_dataset, load_from_disk, Dataset
|
4 |
+
import re
|
5 |
+
from tqdm import tqdm
|
6 |
+
import sys
|
7 |
+
|
8 |
+
def clean_markdown(text):
|
9 |
+
text = re.sub(r'<.*?>','',text)
|
10 |
+
text = re.sub(r'\n+','',text)
|
11 |
+
text = text.replace('#','')
|
12 |
+
return text
|
13 |
+
|
14 |
+
|
15 |
+
def parse_data(ds):
|
16 |
+
"""Parse data into markdown-code pairs"""
|
17 |
+
|
18 |
+
for notebook in tqdm(ds):
|
19 |
+
|
20 |
+
types = notebook["cell_types"]
|
21 |
+
cells = notebook["cells"]
|
22 |
+
|
23 |
+
if len(types)>0:
|
24 |
+
if types[0] == "code":
|
25 |
+
# drop first cell of code to have the notebook start with markdown
|
26 |
+
cells = cells[1:]
|
27 |
+
types = types[1:]
|
28 |
+
#else:
|
29 |
+
# drop first the two cells of markdown followed by code
|
30 |
+
# the first markown cell of a notebook is often a long description of the whole notebook
|
31 |
+
# cells = notebooks["cells"][2:]
|
32 |
+
# types = notebooks["types"][2:]
|
33 |
+
if len(types)>0:
|
34 |
+
if types[-1] == 'markdown':
|
35 |
+
cells = cells[:-1]
|
36 |
+
types = types[:-1]
|
37 |
+
|
38 |
+
if len(cells) % 2 == 0:
|
39 |
+
inner_markdowns = [cells[j] for j in range(len(cells)) if j % 2 == 0]
|
40 |
+
inner_code_snippets = [cells[j+1] for j in range(len(cells) - 1) if j % 2 == 0]
|
41 |
+
|
42 |
+
|
43 |
+
for markdown_block, code_snippet in zip(inner_markdowns,inner_code_snippets):
|
44 |
+
markdown_block = ' '.join([clean_markdown(block[0]) for block in markdown_block])
|
45 |
+
code = '\n'.join([snippet[0] for snippet in code_snippet])
|
46 |
+
output = [snippet[1] for snippet in code_snippet][-1]
|
47 |
+
|
48 |
+
line = {'markdown':markdown_block,
|
49 |
+
'code':code,
|
50 |
+
'output':output,
|
51 |
+
'license':notebook['max_issues_repo_licenses'][0],
|
52 |
+
'path':notebook['max_stars_repo_path'],
|
53 |
+
'repo_name':notebook['max_stars_repo_name'],
|
54 |
+
}
|
55 |
+
yield line
|
56 |
+
|
57 |
+
|
58 |
+
if __name__ == "__main__":
|
59 |
+
file = sys.argv[1]
|
60 |
+
|
61 |
+
dataset = load_dataset("bigcode/jupyter-parsed")
|
62 |
+
with open(file,'w') as out:
|
63 |
+
for line in parse_data(data):
|
64 |
+
out.write(json.dumps(line)+'\n')
|
65 |
+
|
66 |
+
dataset = load_dataset('json',ata_files=file)
|
67 |
+
|
68 |
+
dataset.push_to_hub("bigcode/jupyter-code-text-pairs")
|