README.md

metadata

dataset_info:
  features:
    - name: code
      dtype: string
    - name: language
      dtype: string
    - name: AST_depth
      dtype: int64
    - name: alphanumeric_fraction
      dtype: float64
    - name: max_line_length
      dtype: int64
    - name: avg_line_length
      dtype: float64
    - name: num_lines
      dtype: int64
    - name: task
      dtype: string
    - name: source
      dtype: string
  splits:
    - name: train
      num_bytes: 3081197845.7939653
      num_examples: 1424089
  download_size: 335813854
  dataset_size: 3081197845.7939653
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

This dataset contains deduplicated solution data from TACO dataset, 107k LeetCode problem solutions and an additional 73k CodeForces solutions in Python, Java, C++ from this link.

Filtering strategy:

• AST parsability: we extracted only the AST parsable codes
• AST Depth: only samples with depth from 2.0 to 31.0
• Maximum Line Length: only samples with a maximum of 12.0 to 400.0 characters
• Average Line Length: only samples with 5.0 to 140.0 characters on average
• Alphanumeric Fraction of samples greater than 0.2 and less than 0.75
• Number of Lines: from 6.0 to 300.0
• Language Filter: samples whose docstring scores an English language confidence greater than 99% by the lingua language detector
• Deduplication: samples MinHash deduplicated using a shingle size of 8 and a similarity threshold of 0.8