Upload segment_notebooks.py
Browse files- segment_notebooks.py +57 -0
segment_notebooks.py
ADDED
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import itertools
|
3 |
+
from datasets import load_dataset
|
4 |
+
|
5 |
+
|
6 |
+
def segment_cells(content):
|
7 |
+
|
8 |
+
# segment notebooks into lists of individual cells
|
9 |
+
cells = []
|
10 |
+
cell_types = []
|
11 |
+
for cell in content['cells']:
|
12 |
+
# select only non-empty cells
|
13 |
+
if len(cell['source']) != 0:
|
14 |
+
cells.append(' '.join(cell['source']))
|
15 |
+
cell_types.append(cell['cell_type'])
|
16 |
+
|
17 |
+
return cells, cell_types
|
18 |
+
|
19 |
+
|
20 |
+
def parse_notebook(batch):
|
21 |
+
try:
|
22 |
+
cells, types = segment_cells(json.loads(batch['content']))
|
23 |
+
|
24 |
+
# get cell types and group them into lists
|
25 |
+
cell_type_groups = [list(g) for k,g in itertools.groupby(types)]
|
26 |
+
cell_types = [k for k,g in itertools.groupby(types)]
|
27 |
+
cell_groups = []
|
28 |
+
|
29 |
+
group_start = 0
|
30 |
+
for g in cell_type_groups:
|
31 |
+
cell_groups.append(cells[group_start:group_start+len(g)])
|
32 |
+
group_start += len(g)
|
33 |
+
|
34 |
+
batch['cells'] = cell_groups
|
35 |
+
batch['cell_types'] = cell_types
|
36 |
+
batch['cell_type_groups'] = cell_type_groups
|
37 |
+
|
38 |
+
except:
|
39 |
+
# if json.loads() returns error, skip and add a placeholder
|
40 |
+
batch['cells'] = [['empty']]
|
41 |
+
batch['cell_types'] = ['empty']
|
42 |
+
batch['cell_type_groups'] = [['empty']]
|
43 |
+
|
44 |
+
del batch['content']
|
45 |
+
return batch
|
46 |
+
|
47 |
+
|
48 |
+
if __name__ == "__main__":
|
49 |
+
|
50 |
+
# load dataset
|
51 |
+
dataset = load_dataset("bigcode/the-stack",data_dir="data/jupyter-notebook", split="train",use_auth_token=True)
|
52 |
+
# segment notebooks
|
53 |
+
dataset = dataset.map(segment)
|
54 |
+
# filter out erronous cells via placeholders
|
55 |
+
dataset = dataset.filter(lambda entry: entry['cell_types']!=['empty'])
|
56 |
+
# push to hub
|
57 |
+
dataset.push_to_hub("bigcode/jupyter-parsed")
|