Datasets:

codeset
/

codeset-gym-python

dataset_info:
  features:
    - name: instance_id
      dtype: string
    - name: repo
      dtype: string
    - name: base_commit
      dtype: string
    - name: patch
      dtype: string
    - name: non_code_patch
      dtype: string
    - name: test_patch
      dtype: string
    - name: created_at
      dtype: string
    - name: problem_statement
      dtype: string
    - name: hints_text
      dtype: string
    - name: version
      dtype: string
    - name: environment_setup_commit
      dtype: string
    - name: FAIL_TO_PASS
      dtype: string
    - name: PASS_TO_PASS
      dtype: string
    - name: FAIL_TO_FAIL
      dtype: string
  splits:
    - name: train
      num_bytes: 24352138
      num_examples: 100
  download_size: 8420542
  dataset_size: 24352138
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Codeset-Gym-Python

A gym for training code agents resolve real-world software issues. This version contains issues from Python repositories.

Dataset Structure

Column	Type	Description
`instance_id`	`str`	Formatted as `owner__repo-issueNumber` (e.g. `psf__requests-6091`)
`repo`	`str`	`owner/repo`
`patch`	`str`	Source-code diff that resolves the issue (tests omitted)
`test_patch`	`str`	Diff of test files added/changed by the issue resolving commit
`base_commit`	`str`	Commit hash before the issue resolution
`environment_setup_commit`	`str`	Commit hash of the resolved version, from which the environment was built
`created_at`	`str`	ISO timestamp of the issue resolving commit
`problem_statement`	`str`	Issue title + body
`hints_text`	`str`	Concatenated Issue comments prior to the patch
`version`	`str`	Environment version tags
`FAIL_TO_PASS`	`List[str]`	A json list of strings that represent the set of tests resolved by the PR and tied to the issue resolution.
`PASS_TO_PASS`	`List[str]`	A json list of strings that represent tests that should pass before and after the PR application.

Usage

from datasets import load_dataset

ds = load_dataset("codeset/codeset-gym-python", split="train")

sample = ds[0]
print(sample["problem_statement"])
print(sample["patch"])

The dataset ships as a single train split.

Built by Codeset