pbhogale / simplexgb Goto Github PK

Currently, ordered factors are treated as numbers. Improve this to make them un-normalized integer valued doubles, to reflect the fact that they do not have the same scale as the other numerical values.

Sometimes interpretability is more important than performance. in such cases, having a linear model is important.

1. give an option to use linear model instead of gbm
2. add feature selection : use a small tree based model to calculate feature importance, use important features to construct linear model

Create vignettes using a simple data set showing how to use all the basic workflows.