Datasets:

Maxros
/

test

license: mit
task_categories:
  - text-generation
tags:
  - code
  - java
  - refactoring
  - benchmark
  - software-engineering
size_categories:
  - n<1K
dataset_info:
  features:
    - name: instance_id
      dtype: string
    - name: repo
      dtype: string
    - name: resource_tier
      dtype: string
    - name: repo_stars
      dtype: float64
    - name: repo_forks
      dtype: float64
    - name: language
      dtype: string
    - name: base_commit
      dtype: string
    - name: merge_commit
      dtype: string
    - name: refactoring_types
      list: string
    - name: refactoring_descriptions
      list: string
    - name: file_path
      dtype: string
    - name: source_code_before
      dtype: string
    - name: source_code_after
      dtype: string
    - name: diff
      dtype: string
    - name: is_pure_refactoring
      dtype: bool
    - name: compile_before
      dtype: bool
    - name: compile_after
      dtype: bool
    - name: tests_pass_before
      dtype: bool
    - name: tests_pass_after
      dtype: bool
    - name: FAIL_TO_PASS
      list: string
    - name: PASS_TO_PASS
      list: string
    - name: created_at
      dtype: string
    - name: repo_license
      dtype: string
    - name: lines_changed
      dtype: float64
  splits:
    - name: test
      num_bytes: 279451236
      num_examples: 816
  download_size: 245967508
  dataset_size: 279451236
configs:
  - config_name: default
    data_files:
      - split: test
        path: data/test-*

CS4570 Java Refactoring Benchmark

A benchmark of Java refactoring commits mined from GitHub repositories, split into high-resource and low-resource tiers based on repository popularity. Built for the TU Delft CS4570 ML for Software Engineering course (2026).

Dataset split

Tier	Repositories	Instances	Star range
High-resource	57	618	1000-7298
Low-resource	~60	198	13-90
Total		816

Use the resource_tier field to filter by tier. repo_stars and repo_forks are included for fine-grained popularity analysis.

Construction

Each instance is a single-file refactoring commit that was:

Detected by RefactoringMiner 3.1.3 as containing at least one Java refactoring operation
Verified by building the project before and after the commit using Docker (Maven/Gradle, JDK 17/21)
Filtered to commits where the project compiles and tests pass both before and after (pure refactorings)

High-resource repos have >=1000 stars and >=10 contributors; low-resource repos have <100 stars.

Fields

Field	Type	Description
`instance_id`	string	Unique identifier
`repo`	string	owner/repo identifier
`resource_tier`	string	"high" or "low"
`repo_stars`	int or null	GitHub star count at collection time
`repo_forks`	int or null	GitHub fork count at collection time
`language`	string	Always "java"
`base_commit`	string	Parent commit SHA
`merge_commit`	string	Refactoring commit SHA
`refactoring_types`	list[str]	Refactoring type labels from RefactoringMiner
`refactoring_descriptions`	list[str] or null	Human-readable descriptions (HR only)
`file_path`	string	Java file path that was refactored
`source_code_before`	string	File content at base commit
`source_code_after`	string	File content at merge commit
`diff`	string	Unified diff between before and after
`is_pure_refactoring`	bool or null	True if only refactoring operations changed
`compile_before`	bool or null	Project compiled at base commit
`compile_after`	bool or null	Project compiled at merge commit
`tests_pass_before`	bool or null	All tests pass at base commit
`tests_pass_after`	bool or null	All tests pass at merge commit
`FAIL_TO_PASS`	list[str]	Tests failing before, passing after
`PASS_TO_PASS`	list[str]	Tests passing both before and after
`created_at`	string or null	Commit timestamp (ISO 8601)
`repo_license`	string or null	SPDX license identifier
`lines_changed`	int or null	Lines changed in the refactoring

Refactoring types

Top types across both tiers: Extract Method, Rename Method, Rename Variable, Rename Parameter, Extract Variable, Inline Variable, Change Variable Type, Add Parameter, Add Method Annotation, Rename Attribute.

66 unique refactoring types in total.

Usage

from datasets import load_dataset

ds = load_dataset("Maxros/test", split="test")

# Filter by tier
hr = ds.filter(lambda x: x["resource_tier"] == "high")
lr = ds.filter(lambda x: x["resource_tier"] == "low")

Citation

TU Delft CS4570 -- ML for Software Engineering benchmark, 2026.