Rseslib

General information

Rseslib is a library of machine learning data structures and algorithms implemented in Java. The library is developed by the team from Group of Logic, Faculty of Mathematics, Informatics and Mechanics, University of Warsaw, Poland. The team is headed by Professor Andrzej Skowron.

This web site introduces to the newest version of the library: Rseslib 3. The first version Rseslib 1 started in 1993 and was implemented in C++. Rseslib 2 was the first version of the library implemented in Java and it stands for the core of RSES 2.x.

Rseslib 3 is assumed to provide modular component-based architecture and easy-to-reuse data representations and methods. It is used in TunedIT system for automated evaluation, benchmarking and comparison of data mining and machine learning algorithms.

Download

Rseslib is distributed under GNU GPL license. To download Rseslib and its source code click the links below:

There are also the tools provided for Rseslib:

rsestools-0.9.0.zip

Before running a tool unzip the file. At present there are the following tools available:

Qmak (beta version) - graphical interface for data and classification visualization and for classifier testing and comparison.
To start Qmak run the script qmak.bat.
Simplistic Grid Manager - client-server tool for classifiers testing on many computers in a network.
To start the server run the script sgm-server.bat providing a file with a list of experiments to do, e.g. sgm-server data/experiments.txt.
To start the client run the script sgm-client.bat providing the name or address of a server, e.g. sgm-client localhost.

Qmak is in beta version so there are still the two older graphical interfaces available:

Visual Rseslib - to start Visual Rseslib run the script vr.bat.
Trickster - to start Trickster run the script trickster.bat.

Documentation

The graphical interface Qmak is described in the document qmak.pdf (in Polish). The fundamental programming structures and functions of the library and Simplistic Grid Manager are described in the document rseslib.pdf. The full list of the implemented library components and their location in the library structure is contained in components table.

The older graphical interfaces are described here: visualrseslib.pdf (in Polish) and Trickster documentation (in Polish). More information about Visual Rseslib can be found on Visual Rseslib home page (in Polish).

Preliminary assumptions about the project are collected in the document projekt.ps (in Polish).

Data format

At present the library reads two formats of data tables:

Rseslib, the examples of data tables in this format are available here (small data sets) and here (large data sets); this format consists of two files:
- a header file *.hdr (e.g. mushroom.hdr, census-income.hdr, covtype.hdr), each line is either an empty line, a comment line starting with the '#' character, a missing value line or an attribute line; a missing value line occurs in a header file at most once and it starts with the keyword 'missing_value' that is followed with strings representing a missing value separated with commas and/or white characters; each attribute line starts with an attribute name that is followed with the attribute features separated with commas and/or white characters: the meaning of the attribute (conditional, decision or text), the type of the set of values (numeric or nominal) and optionally with the indicator (skip) if the attribute is to be skipped while loading data;
- a data file *.tab, each line is either an empty line, a comment line starting with the '#' character or a data line; data fields in a data line are separated with a single comma and/or white space characters.
RSES system, the examples of data tables in this format are available here.

Development

The source code is maintained by the SVN system in the repository https://svn.mimuw.edu.pl/repos/rseslib. Access to the source code is available only for people having an account at the server svn.mimuw.edu.pl with permissions to the project rseslib.

For rseslib development Eclipse is recommended. Working with the SVN repository is possible inside Eclipse with the help of Subclipse plugin. The source code convention is assumed to follow Sun Microsystems standard.

All issues on library design and code are discussed at the mailing list RSProject (in Polish).

Contact

Arkadiusz Wojna
email: wojna@mimuw.edu.pl