metadata

dataset_info:
  features:
    - name: prompt
      dtype: string
    - name: reference_code
      dtype: string
    - name: code_context
      dtype: string
    - name: problem_id
      dtype: int64
    - name: library_problem_id
      dtype: int64
    - name: library
      dtype:
        class_label:
          names:
            '0': Matplotlib
            '1': Numpy
            '2': Pandas
            '3': Pytorch
            '4': Scipy
            '5': Sklearn
            '6': Tensorflow
    - name: test_case_cnt
      dtype: int64
    - name: perturbation_type
      dtype:
        class_label:
          names:
            '0': Difficult-Rewrite
            '1': Origin
            '2': Semantic
            '3': Surface
    - name: perturbation_origin_id
      dtype: int64
  splits:
    - name: test
      num_bytes: 3136179
      num_examples: 1000
  download_size: 712717
  dataset_size: 3136179
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*
license: cc-by-sa-4.0
language:
  - code
task_categories:
  - text2text-generation
tags:
  - code-generation
arxiv: 2211.11501

This is a reupload of DS-1000. The metadata dictionary has been extracted into columns and the categorical variables are now ClassLabel types, and the dataset is natively a parquet. The features are as follows:

Column	Type
problem_id	`Value(dtype='int64', id=None)`
prompt	`Value(dtype='string', id=None)`
reference_code	`Value(dtype='string', id=None)`
code_context	`Value(dtype='string', id=None)`
library_problem_id	`Value(dtype='int64', id=None)`
library	`ClassLabel(names=['Matplotlib', 'Numpy', 'Pandas', 'Pytorch', 'Scipy', 'Sklearn', 'Tensorflow'], id=None)`
test_case_cnt	`Value(dtype='int64', id=None)`
perturbation_type	`ClassLabel(names=['Difficult-Rewrite', 'Origin', 'Semantic', 'Surface'], id=None)`
perturbation_origin_id	`Value(dtype='int64', id=None)`

All credits go to the original authors below.

DS-1000 in simplified format

See testing code and more information in the DS-1000 repo.

Reformatting credits: Yuhang Lai, Sida Wang