Optimization Plugin for RapidMiner
The optimization plugin for RapidMiner aims to introduce mathematical optimization based learning capability to RapidMiner. An unconstrained optimization problem, which is called learning (or) training in machine learning can be solved with different algorithms. Each algorithm has a set of configurable items, that are mainly function to be optimized, choice of search direction, regularization type,choice of learning rate (line search) . There are different options or ways available to perform the task pertaining to each configuration. This plugin aims to give a clean separation of these configurable items, so that the user can configure these items before starting the learning. Thus, giving the user, an easily configurable optimization based learner.
Please note, the this is just experimental software and might contain bugs or unfinished operators. If you have any questions, drop me a line. You can find me at:
http://www-ai.cs.uni-dortmund.de/PERSONAL/umaashankar.html venkatesh[dot]umaashankar[at]g mail[dot]com
Just copy the file "rapidminer-Optimization-Extension-1.x.x.jar" to the /lib/plugins folder of your RapidMiner 5 installation. For windows users this could be "C:\Program Files\RapidMiner\lib\plugins" or similar.
Sample Data and Sample Processes
Untar the file optimrepo.tar.gz in your home directory. you will get a directory named "repo". configure it as a rapidminer local repository by pointing to this sample repository, when creating a new repository in Rapidminer. The sample repository contains many sample processes which show case several configurable options of the optimization plugin.
In the current version (1.0.0) of the extension contains the following operators:
Optimization Based Learner: OptimLearner
The optimization plugin uses SLF4j and Logback for logging. A sample logback.xml file is provided in the installJars folder. The Logback configuration file can be specified as shown below while starting the rapidminer
java -Xmx3G -Dlogback.configurationFile=/usr/local/rapidminer/lib/plugins/logback.xml -jar /usr/local/rapidminer/lib/rapidminer.jar
Depending on the loglevel, useful statuses and configuration details are logged. The default log level will be INFO and it does not require any configuration.
Prerequisites: You need a recent version of the RapidMiner 5.x source code. It should be located at "../RapidMiner_Vega" relative to this directory. If this does not apply, change the "rm.dir" property in the build.xml file.
Go for it: Just type
for building the plugin. If you want it packaged, type
and if you are running RapidMiner from your RapidMiner_Vega source directory you can also type
which packages the plugin and places it in "RapidMiner_Vega/lib/plugins".
[Then start RapidMiner.]