How to implement a customized Random Forest classifier I am trying to replicate an extension of Random Forests introduced in a recent research publication for my project. 
For instance, for the binary split of data at each node, instead of randomly assigning all example from each class a binary label, they suggest using an SVM to learn a binary split of data. Furthermore, features at a node are augmented with the decision value of its parent node.
I wanted to know, if there exists an implementation of Random Forest which can be built on? 
 A: I think you need an open-source, readable version of random forests to be able to digg into the code and propose your modifications.
All the basic building blocks you need are available in scikit.learn if you know some python - plus the community is very active and proved to be able to help such efforts in the past. 
A: OpenCV has an implementation of random forests, based on the algorithm of Leo Breiman. Check out the documentation here. It is open source, and fairly readable, active.
A: The obliqueRF packages in R can do discriminative node models based on "ridge regression, partial least squares regression, logistic regression, linear support vector machines, or random coefficients."
That might be a good place to start, as you could probably use the built-in SVM model to partially achieve your goal.  Furthermore, you can take a look at the package source, and modify it to suit your needs.
A: There is a minimal open source random forest in Golang here:
https://github.com/ruffrey/pine
MIT license
