README.md

---
license: mit
size_categories:
- n<1K
task_categories:
- summarization
- text-generation
pretty_name: MCE
dataset_info:
  features:
  - name: class_id
    dtype: string
  - name: class_code
    dtype: string
  - name: skeleton
    dtype: string
  - name: method_code
    dtype: string
  - name: method_summary
    dtype: string
  splits:
  - name: test
    num_bytes: 2166776
    num_examples: 410
  download_size: 381808
  dataset_size: 2166776
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
tags:
- code summarization
- python
- function-level
- class-level
- ClassEval
---

# Modified ClassEval (MCE) Dataset

This dataset is a modification of the ClassEval benchmark, designed for evaluating code summarization models beyond the function level. It explores the impact of function and class contexts on summary quality.  The dataset includes modifications for evaluating at both function and class levels.

Paper: [Code Summarization Beyond Function Level](https://hf.co/papers/2502.16704)

**Dataset Structure:**

The dataset contains samples with the following fields:

* `class_id`: Identifier for the class.
* `class_code`: The code of the class.
* `skeleton`:  The structured blueprint for the class, including class-level and function-level information.
* `method_code`: The code of a method within the class.
* `method_summary`: The summary of the method.


**Dataset Splits:**

* `test`: 410 samples (400 + 10 samples for few-shot learing)

This dataset is part of a broader study investigating code summarization at various levels (function, class, and repository).  The complete study details, code, and evaluation results are available on GitHub: [https://github.com/kilimanj4r0/code-summarization-beyond-function-level](https://github.com/kilimanj4r0/code-summarization-beyond-function-level).