question: building a model from prior data

Is it possible to create the "best" predictive model for a discoveryspace (containing prior knowledge but non-exhaustively sampled) using trim operator without any additional sampling? e.g. a "build best model from prior knowledge" mode. 

What options are required to do this?

