MATSim Code

transform problem into a mono-objective one

also write to file whether step was improving

only one initial value

more configurable

less initial points

do not constrain secondary reduction to be positive

negative reductions might actually make sense, as sampling rate for
secondary round typically higher (threshold includes probability to
meet)

yet another search heuristic

similar to "compass search"

new search heuristic

also gets lost: not always possible to get clustering low or high enough

make sure logged config ends up in config file

use heuristic to generate starting points