Slow KNN Regressor

[cugurm]

Too large evaluator loading time of K Nearest Neighbors regressor (with a lot of training instances, about 100k instances each size of 100 double values).

I now that KNN is a lazy regresor, but are there any eventually speedup?

All the best,
Milan

[vruusmann]

Closing as functional duplicate of #187

The JPMML-Evaluator development branch 1.5.X provides support for choosing an evaluation "approach" based on model configuration. For example, we could use one approach for small all-categorical KNNs and another one for big all-continuous KNNs (as is the case here).

Unfortunately, right now there's only one evaluation "approach" available, which doesn't suit the particular workflow very well.

I'm more likely to implement the speedup in the JPMML-Transpiler library layer (code generation, plus using some helper third-party library).