Skip to content

Clean samples after projection

When projecting samples from uniformly sampled viable space, we have the following independent problems:

  • samples can loose uniform distribution (think: projecting uniformly sampled sphere),
  • samples can become much more dense (think: projecting rectangular box).

This could be fixed by a simple, random elimination of samples until they are again distributed uniformly over the projection domain, with a target coverage density.

Why bother? Because less samples means less model evaluations, i.e. faster projections processing.

This is related to issue #29. HYPERSPACE sampling provides cleaning() function which is used in ELexp() before the actual uniform sampling with MEBS() is done.