Profiling a NetRAX run on a tree dataset

[lutteropp]

I did a profiling run with NetRAX, on a 1000 sites MSA (belonging to a simulated 8 taxon-tree), with 1 random and 1 parsimony starting tree. Turns out that trying arc insertion moves is pretty slow (simply because the 1-arc-insertion-move-neighborhood of a network is quite large), but most runtime is spent in branch-length optimization using that Brent-method (because it does tons of likelihood recomputations).