README.md

metadata

dataset_info:
  features:
    - name: diff
      dtype: string
    - name: is_single_chunk
      dtype: bool
    - name: is_single_function
      dtype: bool
    - name: buggy_function
      dtype: string
    - name: fixed_function
      dtype: string
  splits:
    - name: train
      num_bytes: 16904390254
      num_examples: 656785
  download_size: 5369285762
  dataset_size: 16904390254
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
language:
  - code
pretty_name: megadiff
size_categories:
  - 100K<n<1M

Megadiff, a dataset of source code changes

If you use Megadiff, please cite the following technical report:

"Megadiff: A Dataset of 600k Java Source Code Changes Categorized by Diff Size". Technical Report 2108.04631, Arxiv; 2021.

@techreport{megadiff,
  TITLE = {{Megadiff: A Dataset of 600k Java Source Code Changes Categorized by Diff Size}},
  AUTHOR = {Martin Monperrus and Matias Martinez and He Ye and Fernanda Madeiral and Thomas Durieux and Zhongxing Yu},
  URL = {http://arxiv.org/pdf/2108.04631},
  INSTITUTION = {Arxiv},
  NUMBER = {2108.04631},
  YEAR = {2021},
}