One-at-a-time heuristics
To check which one-at-time (OAT) parameters projections are viable (in getViableProjections
), instead of projecting coordinates in each single sample of parameters values, rather check whether across all samples there exists a coordinate value close to the projection value.
Intuitively for parameters with wide sampling intervals and projected to zero or epsilon values, having samples with corresponding coordinate values close to the lower boundary of the parameter box bmin
is an indication for possible viable projections of these parameters.
Initially added in commit #cb41dfa9 via useOATheuristics
option. Currently implementation is very crude, and checks proximity to the bmin
value (fixed n
-fold proximity, with n=3
). It doesn't make sense for parameters other than ones described in previous paragraph. Improve implementation to include projection value proximity concept explicitly, and test how efficient it is.