Open stumarcus314 opened 2 months ago
With the default mesh and direction type used by Nomad, binary variables are converted into integer variables (granularity 1) with 0 and 1 bounds. Other direction types maybe developed in the future that handle binary differently.
If the variables in my optimization problem are all binary, is there any difference internally to NOMAD between specifying them as binary by setting BB_INPUT_TYPE to (B B B) versus specifying them as integers by setting BB_INPUT_TYPE to (I I I) with lower bounds 0 and upper bounds 1? For example, does specifying them explicitly as binary using the first approach realize computational and/or memory efficiencies for NOMAD?