Another common situation in process control is that a particular output of interest is a function of a large number of variables, but the relationship to many of them is simple (globally linear, for example). We have already seen how Vizier can mix attributes that have globally linear effects with those that have complex non-linear effects. It is done by selecting input weights of 0 in the distance metric for the attributes that have globally linear effects, and non-zero weights for those with significant non-linearities. In fact, many of the applications shown in fig. 17 have this property.