Skip to content

v0.10.0

Compare
Choose a tag to compare
@fairbrot fairbrot released this 02 Oct 11:21
· 136 commits to master since this release
fafb5b8
  • Introduced sparse approximation methods
    • Subset of regressors, deterministic training conditional, fully independent training conditional and full-scale approximation are all available
  • Extended functionality to include leave-one-out cross-validation
  • Introduced functionality to enable variational inference in GPs with non-Gaussian data.
    • The approach used is a variant of that presented in Khan et. al.
    • Currently limited to only Poisson data. Additional likelihood functionality will be added soon.
  • Deprecated GPMC in favor of GPA. This is to be inline with that fact that approximate inference in a GP is not limited to MCMC, but variational methods can now be used.