Some days ago I watched this lecture by Nando De Freitas on Decision Trees so I decided to try my personal implementation of the algorithm (also to learn how to use cmake). It supports categorical and continuous variables as descriptors. The labels are not constrained on two levels only. It does not rely on any external library.

The criterion to select the best variable to split the dataset is based on the achievement of the maximum information gain calculated on entropy.

In order to build you can simply access the project root directory and type

mkdir build 
cd build 
cmake ..
make 

For the moment it is only possible to build the trees. I will provide error estimates in the near future.