README.md

metadata

dataset_info:
  features:
    - name: code_before
      dtype: string
    - name: edits
      dtype: string
    - name: next_edit
      dtype: string
    - name: code_after
      dtype: string
    - name: label_window
      sequence: string
    - name: instruction
      dtype: string
    - name: html_url
      dtype: string
    - name: file_name
      dtype: string
  splits:
    - name: train
      num_bytes: 12415666705
      num_examples: 2633966
    - name: validation
      num_bytes: 1314363212
      num_examples: 376282
    - name: test
      num_bytes: 2432871766
      num_examples: 752565
  download_size: 2729492743
  dataset_size: 16162901683
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*

Dataset Card for Dataset Name

Processed version of the code-philia/CoEdPilot-generator for the next edit prediction task from the paper CoEdPilot.

Dataset Structure

• code_before - initial code to be modified
• edits - previous edits
• next_edit - next edit to be done
• code_after - code after the edit
• label_window - labels of happened with each line in the code before. There are three options: keep (nothing changes), add (adds new content after the line), replace (replaces all lines with new content).
• instruction - presumably commit message from the commit that was used for a datapoint
• html_url - commit url
• file_name - changed file

Reproduction script

https://github.com/JetBrains-Research/comment_resolution/blob/main/notebooks/coedpilot_dataset.ipynb