Random sampling of parameter space

[cerebis]

A lingering thought, grid search has been shown to be poor means of exploring high dimensional spaces when the objective surface is not a simple convex shape. Complex response surfaces are probably more common than not.

We could easily implement random sampling. Define valid ranges of parameters and the number of samples.

This can actually better highlight the effort required to explore a hyper-surface. The sense of sparsity is more pronounced.

[koadman]

something smarter than grid sweep would be rad