The library is header-only (and heavily relies on C++20 templates and concepts), so just include the appropriate files when compiling.

Features

In cartpp, one will find an implementation of CART for regression and classification trees. The existing loss functions so far are:

• MSE
• Poisson deviance
• Negative binomial deviance

In addition to classical CART, cartpp proposes an implementation of Gini-trees (TODO: add link to paper). That is, in addition to a construction of trees where the loss at each leaf is independent of the loss at every other leaf; cartpp proposes a loss at each leaf that depends on the whole tree (the distribution of predictions). That loss is the Gini index of the tree.

Bindings

Python binding

The module pycart is a Python binding using Cart++. The binding is written using Cython.

R binding

Will come at some point (maybe?)

TODO

See TODO.md