Changelog
V 1.6- extended and refactored interface of icl_base to allow for complete setup of learning algorithms and model structures
- added possibility to use IRMA with zero stiffness
- fixed minor bug in IRMA initialization
- added a new drift dataset for changing targets
V 1.5
- added more learning algorithms: Perceptron, IRMA
- extended GLTs to be of any dimensionality and to arbitrarily partition the inputs
- added a new 1D polynomial and 3D linear dataset
V 1.4
- changed main function to be callable from another script and making it executable on its own at the same time
- added more learning algorithms: PA-I, PA-II, AROW, GH (with full, exact, drop, project variants), CW
V 1.3
- added support for saving parameter development
- added option to choose learning algorithm through a variable method - the learning algorithm is identified by adding the method-identifier to icl_initILS_ and icl_learn_
- included PA and RLS as learning algorithms
V 1.2
- added initialization routine for ILS - now every learning method can initialize further parameters in this routine
- added two new generators: relearning and spiral loop
V 1.1 b
- fixed polynomial generator, now only one offset term is added to the phi-vector regardless of the dimension
V 1.1 a
- fixed downwards compatibility of seeding of random number generator, using the old style seeding now to achieve exactly the same results in every Matlab version
V 1.1
- fixed crossedridge-generator to generate output values in [-1 1]
- added two new generators: highdimlin, highdimnonlin
- fixed random number seed to be the same at every run for rand and randn
- added examples for structural initialization
- added examples for parameter initialization
V 1.0 a
- added common basis of approximator
V 1.0
- initial framework release