The optimization branch

[theadityasam]

The link for benchmarks: https://theadityasam.github.io/iregbenchmark/benchmarks.html

I encountered few issues which I'll list over here:

• For prostate, a weird thing happens, for some particular dataset sizes the runtimes reduce drastically(by x1000 times!), hence the plot looks very skewed. Strangely this happens only when iregnet optimised is added to the benchmarking tests; not when benchmark is performed on master iregnet and glmnet
• For 100000 observations, we see a pattern we expect, glmnet stays consistent whereas iregnet and iregnet optimised are linear with the optimised branch being faster
• penalty.learning produces NAN errors in the optimised branch but not in the master branch
• While trying with neuroblastoma, iregnet works fine but optimised branch produces the following error:

  error: Row::subvec(): indices out of bounds or incorrectly used
  Error in fit_cpp(x, y, family, alpha, lambda_path = lambda, num_lambda = num_lambda,  : 
  Row::subvec(): indices out of bounds or incorrectly used