Speed up statsample-glm

[lokeshh]

Statsample-GLM currently takes up a lot of time to fit models. For datasets having 1000 rows it can take hours to fit the model.

[v0dro]

What do you suggest should be done? Let's first profile the code and see the slowest places. Then we can optimize these and see if we should write C extensions or port things to NMatrix.

[envp]

I'm interested in this, where should I start?

[agisga]

I think that maybe one way to start would be to look at how things are done in other tools such as R and Python.
If it helps, I recently found these notes on high-performance GLM solvers (in particular, on variants of the IRLS algorithm): http://bwlewis.github.io/GLM/ (the example codes are in R, but it should be easy to translate them into Ruby).

[envp]

Thanks! I'll go through the recommended reading for now. Have a good deal of schoolwork coming up right now.

I'll report back ASAP!