Skip to content
muuki88 edited this page Apr 30, 2012 · 7 revisions

Situation

Every user produces more and more data. There are some data mining frameworks or tools, e.g WEKA or RapidMiner, which help to gain valuable knowledge out of this data. Both applications based on Java don't make use of modular technology like OSGi, which we see as a big shortcoming.

Goals

We don't want to provide a new data mining framework which should replace the olds. Instead we want to add an abstraction layer above existing frameworks.

  • Easy deployment to OSGi environments
  • Easy integration into existing systems via a Service Orientated Architecture (SOA)
  • Scala API for more powerful and expressiv code (along with Java API)
  • Highly parallel execution via Actor based programming with Akka
  • Wrappper data mining API for existing libraries (e.g. WEKA)
  • Developer tools to easly generate portable data mining tasks

Use Cases

There are two typical use-cases:

  1. Use Knowing as a framework for consumer applications
  2. Implement / migrate algorithms to use them in OSGi backend environments

Consumer applications

You want to create a modular data mining application? Knowing helps you to plugin in new algorithms by using a OSGi architecture. Algorithms are just services and you can register and deregistern them. This keeps your application small and flexible. You can also run the algorithms on a java application server which means you can use fancy UI-Systems like Vaadin to provide data mining tasks right on your website.

Implement / migrate algorithms

Knowing offers an API which makes it easy to migrate your old algorithms or implement new ones. Currently there's only a WEKA wrapping API which wraps existing WEKA classifier, clusterer and filter into services which can be used right away.

Building and Running

At the moment Knowing is ALPHA stadium. However it actually works. To build and run all Knowing plugins your have following requirements:

  • java 1.6
  • scala 2.9.0.1
  • OSGi Environment (Equinox, Felix,..)
  • Google Guava

Checkout the source and open it with your favorite Java IDE. Take care that the Bundle-Manifest is include when building. Start the OSGi Framework or create an OSGi-launchconfiguration in Eclipse and install/start the following bundles:

The plugin in knowing.test includes examples how to run example configurations. knowing.test is an eclipse RCP application. You can build it with the following dependencies

  • de.lmu.ifi.dbs.knowing.core.swt
  • de.lmu.ifi.dbs.knowing.core.swt.charts

Eclipse Plugin

There is a small Editor for Eclipse Helios which let's you define your own DPUs. However it is not yet released, because some features must be completed first.