Miscellaneous notes

[jamesrobertlloyd]

Hessian delta of 1e-6 will not always be appropriate e.g. location parameters for unscaled data

[jamesrobertlloyd]

The default prior in the laplace approx needs to be revisited

[jamesrobertlloyd]

NaNs for scores mess up sorting routines - affects the printing of results I think