Implement 2D and 3D learners

Same as 1D, but sampling from a higher-dimensional parameter space in a way that is not reduced to consecutive samples from 1D cuts (although initially this may be a good strategy).