jimneussl's picture
Update README.md
c529098 verified
metadata
dataset_info:
  features:
    - name: id
      list: int64
    - name: title
      dtype: string
    - name: text
      dtype: string
    - name: target
      dtype: string
    - name: ctxs
      list:
        - name: text
          dtype: string
        - name: title
          dtype: string
    - name: primary_domain
      dtype: string
  splits:
    - name: train
      num_bytes: 48308525
      num_examples: 12041
  download_size: 10808808
  dataset_size: 48308525
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Dataset Description

This dataset is a filtered version of the wellecks/naturalproofs-gen dataset. The original dataset contains mathematical proofs from ProofWiki. This filtered version removes entries that contain URLs, Wiki file links, or markdown images in either the text (problem description) or target (proof) fields.


Features

  • id: A list representing the unique identifier of the entry.
  • title: The title of the mathematical proof.
  • text: The problem description or statement of the theorem to be proved.
  • target: The mathematical proof itself.
  • ctxs: A list of dictionaries containing contextual information (e.g., related theorems or definitions).
  • primary_domain: The primary mathematical domain of the entry (e.g., Algebra, Calculus, Geometry, Number Theory, Logic, Combinatorics, Inequalities).

Motivation

The dataset was created to provide a cleaner version of the NaturalProofs data, suitable for tasks where the presence of external links or images might be undesirable.