Nonparametric geometries


Is it possible to train a network on a set of nonparametric geometries? For example, train the network on many different nonparametric geometries so I could use an arbitrary geometry on inference.

Thanks in advance

Hi @gorpinich4

Does the DiscreteGeometry component work for your need? We have an example to make a set of tessellated geometry such that it looks parametric (you can define define the parameterization as some arbitrary variable).

In the constraint, this will sample one of the provided geometries in a discrete list, than sample a point from that geometry.