utility functions for logistic regression and ols

[bcjaeger]

default methods to find linear combos of predictors in classification and regression trees

[bcjaeger]

This issue could theoretically be closed, as logreg_fit and linreg_fit in utility.cpp seem to work well enough, but a few things about logreg_fit could be fleshed out more. In particular, could this function run faster if we used the L-BFGS algorithm? And, could we go one step further and use L-BFGS to run penalized regression? I don't know much about this area other than it appears there are libraries for it in C++ and they may work with armadillo.

@ciaran-evans, would you be interested in seeing if it would be feasible to write a logistic regression function using L-BFGS that runs faster than the one in aorsf? The new function could just be a stand alone .cpp file at first and we could work through pulling it into aorsf later. No pressure...this is a loosely defined problem, and it's perfectly alright if you would like to work on a different thing. One potential pro is that this is not time-sensitive.