jamesrobertlloyd / gp-structure-search

MIT License
229 stars 83 forks source link

Miscellaneous notes #48

Open jamesrobertlloyd opened 11 years ago

jamesrobertlloyd commented 11 years ago

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

jamesrobertlloyd commented 11 years ago

The default prior in the laplace approx needs to be revisited

jamesrobertlloyd commented 11 years ago

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