this is sick - the google linear programming system.
https://developers.google.com/optimization/lp/glop
prior to glop, i had already tried scipy.optimise.minimise (SLSQP) and scipy.optimise.linprog (simplex). unfortunately with large group sizes (> 200 variables) SLSQP took > 25 minutes just to solve each group, and linprog (somehow) just returned an error message. for smaller sized group both performed well though (simplex was slightly faster).
enter glop - what took SLSQP > 25 minutes was almost instantaneous in glop!!!
sick. i finished running my 35,000 groups (from variable sizes ranging up to a few thousand) in around 30 minutes, locally (on my not-that-good work laptop with other stuff running in the background).
my colleague mentioned running the same set of data on SASGrid takes about 15 minutes.
glop is pretty sick!! no grid computing required lol.
everyone, PLEASE USE IT.