NEW! Debellor is being replaced now by Data Pipes, a Python framework for scalable processing of Big Data and Data Streams, more mature and functional than Debellor.
Debellor is an open source framework for scalable data mining and machine learning. You can implement your own algorithms in Debellor's architecture and combine them with pre-existing ones to create complex data processing networks and experimental setups. The unique feature of Debellor is data streaming, which enables efficient processing of massive data and is essential for scalability of algorithms. Debellor is used in TunedIT system for automated evaluation, benchmarking and comparison of learning algorithms.
Latest release can be downloaded here.
Debellor simplifies implementation of complex yet efficient algorithms.
Debellor is written in Java and distributed under GNU General Public Licence. Hosted by SourceForge.net.
- 120+ algorithms are already available and can be used as building blocks. These include all classifiers from Weka and Rseslib libraries, all filters from Weka and a reader of arff files.
- Simplicity. All algorithms are accessible through the same simple interface of a Cell.
- Scalability. Thanks to stream architecture algorithms may process data on the fly instead of keeping all of them in memory. This enables efficient handling of large volumes of data and gives you freedom of designing arbitrarily complex processing networks.
- Extendibility. You may define new data types, specific to your application domain. See DataObject and DataType.
- Multithreading. Take full advantage of a multi-core CPU with parallel execution of experiment. Debellor takes care of thread management and synchronization, you only decides where to make a splitting point between threads.
- Javadoc - start reading from Cell class.
- Presentation & paper for Debellor 0.6.2. Note that in the latest release of Debellor names of some classes or methods may be different.
- FAQs on the Wiki. You are welcome to edit, for example add new questions.
Debellor 1.0 can be downloaded here. The file contains Rseslib and Weka libraries, as well as example code (from org.debellor.example.Main in debellor.jar). Requires Java Runtime Environment (JRE) 1.6 or later.
To run the example, unzip the downloaded file and execute example.bat (on Windows) or example.sh (Linux). Note that on Linux you first have to set execution flag of the script, for example:chmod +x example.sh
Currently Debellor has no GUI, so you can use it only as a library of algorithms and a framework which defines base classes for your own algorithms (Cell) and data types (DataObject, DataType). Remember to put
debellor.jar, plus optionally
rseslib.jar(bundled in Debellor distribution) on CLASSPATH when compiling and executing your code. If you use IDE, for example Eclipse, you must include these JARs in dependencies for your project (in Eclipse, click Project -> Properties -> Java Build Path -> Libraries).
History of releases
- 30.07.2009: Debellor 1.0
- 19.02.2009: Debellor 0.6.2
- 31.10.2008: Debellor 0.6.1
- 27.09.2008: Debellor 0.6
- 15.07.2008: Debellor 0.5
In next releases
- Multi-input and multi-output cells
- Composite cells (e.g. meta-learning)
- Distributed execution of algorithms
- And more ...
Debellor is developed by Marcin Wojnarski from Warsaw University, Faculty of Mathematics, Informatics and Mechanics, in cooperation with the research group led by Prof. Andrzej Skowron.
Please send all comments, questions, bugs... to: mwojnars (at) users.sourceforge.net
Your feedback is very helpful in further development of Debellor.