I'm very excited about our next project because it includes very useful techniques for model selection and analysis. I'm also excited about adding the Go Grammar to the built-in programming languages of GeneXproTools. Below I outline the major features of this new project:
Being able to select the very best model from all the generated models is a crucial component of model design, with cross-validation at the heart of it. With this new project we will be implementing cross-validation in a way that is fully automatic and you just have to choose the fold and the metric. We will be implementing cross-validation for all Fitness Functions and Favorite Statistics in the Regression Framework, Classification & Logistic Regression.
Hits/Outliers Favorite Statistics
The Hits/Outliers statistics that are now only available when certain fitness functions are selected, will be extended to all fitness functions of the Regression Framework and Time Series Prediction. They'll have adjustable parameters both for the error type (absolute or relative) and error value. This way you'll be able to explore the new features introduced with Mini-Release 1 (especially multi-class classification in the Regression Framework) using all kinds of fitness functions and then easily evaluate the accuracy of your models.
The Variable Importance Chart is an essential tool in model evaluation and analysis and GeneXproTools users tend to use it a lot. So we are bringing this chart more to the front both through a readily accessible icon and a menu.
The Go Grammar, the amazing gift from Glenn Lewis to all GeneXproTools users a while back (see the forum discussion), will be added to the built-in programming languages of GeneXproTools with this mini-release. We will be building on Glenn's contribution (both the Math Grammar and all the Boolean Grammars) and also adding the new 39 math functions that were introduced with MR1. Again, many thanks to Glenn for his contribution!
When we introduced the R Grammar in GeneXproTools 5.0, we weren't sure about the usefulness of including also the Boolean Grammars and so we left them out. But users have been asking for them so we decided to add the R Boolean Grammars with this mini-release.
As with the previous project "New Project: Multi-class Classification & Trading Strategies", we will be blogging actively about this one too. So please join us and add your voice to the discussions!