Efficient Implementation of Decision Forests
J. Shotton, D. Robertson, and T. Sharp
This chapter describes a variety of techniques for writing efficient, scalable, and general-purpose decision forest software. It will cover:-
- Algorithmic considerations, such as how to train in depth first or breadth first order;
- Optimizations, such as cheaply evaluating multiple thresholds for a given feature;
- Designing for multi-core, GPU, and distributed computing environments; and
- Various `tricks of the trade', including tuning parameters and dealing with unbalanced training sets.
Springer-Verlag London 2013