RSSalg-software

A tool for co-training application on single-view datasets. RSSalg software is intended for easy and flexible experimenting with both RSSalg and co-training.

Co-training is a major semi-supervised learning technique. It is applicable on the datasets whose features can be divided in two separate feature subsets called views. Ideally, this feature division is dictated by two independent sources of information describing the data.

There are many real-world problems where the labeled data is scarce and thus would benefit from co-training application, but we lack the knowledge of the a natural way to separate the features in two views.

Random Split Statistic algorithm (RSSalg) is a co-training based algorithm that can be applied on single-view datasets. For a detailed explanation of RSSalg, please refer to:

J. Slivka, A. Kovačević and Z. Konjović: "Combining CoTraining with Ensemble Learning for Application on Single-View Natural Language Datasets", Acta Polytechnica Hungarica, 10.2, 133-152, 2013

Getting Started

These instructions will get you a copy of the project up and running on your local machine for development and testing purposes.

The illustrative example and detailed instructions of running RSSalg software is located in /docs/IllustrativeExample.pdf

Prerequisities

In order to run RSSalg software you need to have the following things installed

Java v1.7 or above

To install Java go to http://www.oracle.com/technetwork/indexes/downloads/index.html#java

Building the project

The easiest way to build and run the project is by using Apache Ant (http://ant.apache.org/). You will find the Apache Ant script named "build.xml" in the root of the project. After installing Ant, open the terminal and navigate to the root of the project. Type the command ant. This will compile and package the code in an executable located in /dist/jar/RSSalg.jar

Optionally, you load the source code from the src folder into your favorite IDE. Note that RSSalg software implementation depends on Weka (version 3.7.0). Thus, you will need to add /lib/weka.jar library into your project. The starting point for running RSSalg software is the class StartExperiment, located inside src/application.

Running the project by using the built executable

Open the terminal and navigate to the executable RSSalg.jar.

1. Running RSSalg software in console:

   java -jar RSSalg.jar <properties_folder> <experiment_properties>
   

   where

   • properties_folder is the folder containing the following property files (needed for execution of all experiments):
     • data.properties - parameters describing the dataset used in the experiment
     • cv.properties - parameters for setting the X-fold-cross validation experiment
     • co-training.properties - parameters of the underlying co-training algorithm
     • GA.properties - properties of the genetic algorithm used in RSSalg threshold optimization
   • experiment_properties is the property file containing the containing the desired settings for concrete experiment that should be run

2. Running RSSalg software as swing application

   java -jar RSSalg.jar
   

Running the project by using Ant

Open the terminal and navigate to the root of the project (containing build.xml):

• You can run RSSalg software in GUI mode by typing the command ant run_GUI

• You can run RSSalg software in console mode by typing the command ant run_console. In this case you need to modify the arguments of target run_console located in build.xml to correspond to the adequate properties_folder and experiment_properties.

Author

• Jelena Slivka - GitHub

License

This project is licensed under the GNU General Public License.