The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets
library). If this is not possible, please
open a discussion
for direct help.
Dataset Card for CodeSearchNet corpus
Dataset Summary
CodeSearchNet corpus is a dataset of 2 milllion (comment, code) pairs from opensource libraries hosted on GitHub. It contains code and documentation for several programming languages.
CodeSearchNet corpus was gathered to support the CodeSearchNet challenge, to explore the problem of code retrieval using natural language.
Supported Tasks and Leaderboards
language-modeling
: The dataset can be used to train a model for modelling programming languages, which consists in building language models for programming languages.
Languages
- Go programming language
- Java programming language
- Javascript programming language
- PHP programming language
- Python programming language
- Ruby programming language
Dataset Structure
Data Instances
A data point consists of a function code along with its documentation. Each data point also contains meta data on the function, such as the repository it was extracted from.
{
'id': '0',
'repository_name': 'organisation/repository',
'func_path_in_repository': 'src/path/to/file.py',
'func_name': 'func',
'whole_func_string': 'def func(args):\n"""Docstring"""\n [...]',
'language': 'python',
'func_code_string': '[...]',
'func_code_tokens': ['def', 'func', '(', 'args', ')', ...],
'func_documentation_string': 'Docstring',
'func_documentation_string_tokens': ['Docstring'],
'split_name': 'train',
'func_code_url': 'https://github.com/<org>/<repo>/blob/<hash>/src/path/to/file.py#L111-L150'
}
Data Fields
id
: Arbitrary numberrepository_name
: name of the GitHub repositoryfunc_path_in_repository
: tl;dr: path to the file which holds the function in the repositoryfunc_name
: name of the function in the filewhole_func_string
: Code + documentation of the functionlanguage
: Programming language in whoch the function is writtenfunc_code_string
: Function codefunc_code_tokens
: Tokens yielded by Treesitterfunc_documentation_string
: Function documentationfunc_documentation_string_tokens
: Tokens yielded by Treesittersplit_name
: Name of the split to which the example belongs (one of train, test or valid)func_code_url
: URL to the function code on Github
Data Splits
Three splits are available:
- train
- test
- valid
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
All information can be retrieved in the original technical review
Corpus collection:
Corpus has been collected from publicly available open-source non-fork GitHub repositories, using libraries.io to identify all projects which are used by at least one other project, and sort them by “popularity” as indicated by the number of stars and forks.
Then, any projects that do not have a license or whose license does not explicitly permit the re-distribution of parts of the project were removed. Treesitter - GitHub's universal parser - has been used to then tokenize all Go, Java, JavaScript, Python, PHP and Ruby functions (or methods) using and, where available, their respective documentation text using a heuristic regular expression.
Corpus filtering:
Functions without documentation are removed from the corpus. This yields a set of pairs ($c_i$, $d_i$) where ci is some function documented by di. Pairs ($c_i$, $d_i$) are passed through the folllowing preprocessing tasks:
- Documentation $d_i$ is truncated to the first full paragraph to remove in-depth discussion of function arguments and return values
- Pairs in which $d_i$ is shorter than three tokens are removed
- Functions $c_i$ whose implementation is shorter than three lines are removed
- Functions whose name contains the substring “test” are removed
- Constructors and standard extenion methods (eg
__str__
in Python ortoString
in Java) are removed - Duplicates and near duplicates functions are removed, in order to keep only one version of the function
Who are the source language producers?
OpenSource contributors produced the code and documentations.
The dataset was gatherered and preprocessed automatically.
Annotations
Annotation process
[More Information Needed]
Who are the annotators?
[More Information Needed]
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
[More Information Needed]
Licensing Information
Each example in the dataset has is extracted from a GitHub repository, and each repository has its own license. Example-wise license information is not (yet) included in this dataset: you will need to find out yourself which license the code is using.
Citation Information
@article{husain2019codesearchnet, title={{CodeSearchNet} challenge: Evaluating the state of semantic code search}, author={Husain, Hamel and Wu, Ho-Hsiang and Gazit, Tiferet and Allamanis, Miltiadis and Brockschmidt, Marc}, journal={arXiv preprint arXiv:1909.09436}, year={2019} }
Contributions
Thanks to @SBrandeis for adding this dataset.
- Downloads last month
- 7,308