This tutorial re-visits DSIVC or “data space inversion with variable control”. DSIVC is a numerically cheap way to undertake optimization under uncertainty. The present tutorial continues a previous tutorial on the same subject by presenting a new (and somewhat more efficient) methodology for achieving similar optimization outcomes.
The tutorial also examines circumstances that may challenge data space inversion (i.e. DSI) in general and DSIVC in particular. Actually, these circumstances can also challenge ensemble methods more broadly. They occur when predictive probability distributions are extremely bimodel, with sharp peaks at either end of a limited predictive range. This situation is not uncommon where particle counts comprise predictions of management interest.