astminer
A library for mining of path-based representations of code and more, supported by the Machine Learning Methods for Software Engineering group at JetBrains Research.
Supported languages of the input:
- Java
- Python
- C/C++
- Javascript (beta) (see issue)
Version history
See changelog
About
Astminer was first implemented as a part of pipeline in the the code style extraction project and later converted into a reusable tool.
Currently it supports extraction of:
- Path-based representations of files
- Path-based representations of methods
- Raw ASTs
Supported languages are Java, Python, C/C++, but it is designed to be very easily extensible.
For the output format, see the section below.
Usage
Use as CLI
- Run
./gradlew shadowJar
in project directory - Now you can use shell script to run cli
./cli.sh optionName parameters
, whereoptionName
is one of the following options:
Preprocess
Run preprocessing on C/C++ project to unfold #define
directives.
In other tasks, if you feed C/C++ file with macroses, they will be dropped as well as their appearances in code.
./cli.sh preprocess --project path/to/project --output path/to/preprocessedProject
Parse
Extract ASTs from all the files in supported languages.
./cli.sh parse --lang py,java,c,cpp --project path/to/project --output path/to/result --storage dot
PathContexts
Extract path contexts from all the files in supported languages and store in form fileName triplesOfPathContexts
.
./cli.sh pathContexts --lang py,java,c,cpp --project path/to/project --output path/to/results --maxL L --maxW W --maxContexts C --maxTokens T --maxPaths P
Code2vec
Extract data suitable as input for code2vec model.
Parse all files written in specified language into ASTs, split into methods, and store in form method|name triplesOfPathContexts
.
./cli.sh code2vec --lang py,java,c,cpp --project path/to/project --output path/to/results --maxL L --maxW W --maxContexts C --maxTokens T --maxPaths P --split-tokens --granularity method
Integrate in your mining pipeline
Import
Astminer is available in Bintray repo. You can add the dependency in your build.gradle
file:
repositories {
maven {
url "https://dl.bintray.com/egor-bogomolov/astminer"
}
}
dependencies {
compile 'io.github.vovak.astminer:astminer:0.6'
}
If you use build.gradle.kts
:
repositories {
maven(url = "https://dl.bintray.com/egor-bogomolov/astminer/")
}
dependencies {
compile("io.github.vovak.astminer", "astminer", "0.6")
}
Examples
If you want to use astminer as a library in your Java/Kotlin based data mining tool, check the following examples:
-
A few simple usage examples can be run with
./gradlew run
. -
A somewhat more verbose example of usage in Java is available as well.
Please consider trying Kotlin for your data mining pipelines: from our experience, it is much better suited for data collection and transformation instruments.
Output format
For path-based representations, astminer supports two output formats. In both of them, we store 4 .csv
files:
node_types.csv
contains numeric ids and corresponding node types with directions (up/down, as described in paper);tokens.csv
contains numeric ids and corresponding tokens;paths.csv
contains numeric ids and AST paths in form of space-separated sequences of node type ids;path_contexts.csv
contains labels and sequences of path contexts (triples of two tokens and a path between them).
If the replica of code2vec format is used, each line in path_contexts.csv
starts with a label,
then it contains a sequence of space-separated triples. Each triple contains start token id, path id, end token id, separated with commas.
If csv format is used, each line in path_contexts.csv
contains label, then comma, then a sequence of ;
-separated triples.
Each triple contains start token id, path id, end token id, separated with spaces.
Other languages
Support for a new programming language can be implemented in a few simple steps.
If there is an ANTLR grammar for the language:
- Add the corresponding ANTLR4 grammar file to the
antlr
directory; - Run the
generateGrammarSource
Gradle task to generate the parser; - Implement a small wrapper around the generated parser. See JavaParser or PythonParser for an example of a wrapper.
If the language has a parsing tool that is available as Java library:
- Add the library as a dependency in build.gradle.kts;
- Implement a wrapper for the parsing tool. See FuzzyCppParser for an example of a wrapper.
Contribution
We believe that astminer could find use beyond our own mining tasks.
Please help make astminer easier to use by sharing your use cases. Pull requests are welcome as well. Support for other languages and documentation are the key areas of improvement.
Citing astminer
A paper dedicated to astminer (more precisely, to its older version PathMiner) was presented at MSR'19. If you use astminer in your academic work, please cite it.
@inproceedings{kovalenko2019pathminer,
title={PathMiner: a library for mining of path-based representations of code},
author={Kovalenko, Vladimir and Bogomolov, Egor and Bryksin, Timofey and Bacchelli, Alberto},
booktitle={Proceedings of the 16th International Conference on Mining Software Repositories},
pages={13--17},
year={2019},
organization={IEEE Press}
}