Creator:Henk-Jan Lebbink
Function:Find functions that describe variables in datasets (classification)
Input Type:xlsx, discrete
Input from Agent:Dizzy, Data Selection Agent
Output Type:Classification advice
Output to Agent:Advice
Short Description:

The Rikku agent allows us to find non-linear functions that can be used to predict the values of output variables from input variables.  The Rikku will be trained with historic data, this training stores the relations between values of variables such that the Rikku has the ability to generalize.

This process of storing is colloquially referred to as “learning”. For example, a physician may have diagnosed thousands of patients, they were all unique in the sense that their symptoms were different, yet the physician could diagnose without an unambiguous instruction booklet. The physician can be said to have the ability to generalize over historic data. The Rikku does precisely that, after internalizing historic data, it yields a function that can be used to predict new situations that were not part of the historic train data.
Learning is a computationally expensive task, and similar to humans, we want to start learning to predict some output variable by first looking at those variables that have the most information in common with the output. Another agent, the Lenny, can perform this task of finding the most
useful variables.

Technically, Rikku is an implementation of a restricted Boltzmann machine that can perform machine learning using contrastive divergence. 

© 2015 Alan Turing Institute Almere