Currently I am using matlab to do my research, I use the Gradient Boost Regression Tree (GBRT). I want to move my research to production, the run time requirement is it needs to be quick.
My understanding is that matlab GBRT is implemented using java and my testing have shown it is not fast. I think an open-source C++ might help.
The package should be able to persist the fitting, so at the run time it will load the previous saved fitting and only do prediction, so fitting time is not that critical. I only saw OpenCV mean my requirement C++/GBRT, anybody has experience with it and other suggestions?