Hi,
Are there any experiences using PyMC3 with random forests? There are three cases I am interested in.
- Using random forest kernels within a Gaussian Process see for instance here https://arxiv.org/abs/1402.4293. Briefly, a random forest kernel is derived from the similarity among observations derived from fitting a random forest. So it involves fitting a RF and deriving a similarity matrix among observations.
- The second one is using bayesian random forests, for instance see http://ieeexplore.ieee.org/document/6920043/
- and the third one is optimizing the parameters of the random forest (e.g., depth) using PyMC3. For instance it can be done with GPyOpt . this may be relevant for the point 1 since ideally we want optimise the parameters of the kernel (random forest kernel).