v0.5

What's Changed

• Transformations by @daniel-dodd in #109
• Natgrads by @daniel-dodd in #90
• Fix bug in Matern12 kernel by @thomaspinder in #119
• Intro to GP notebooks by @thomaspinder in #117
• Add verbose option. by @daniel-dodd in #116
• Numpyro by @thomaspinder in #122
• Distrax reversion by @thomaspinder in #125
• Kernel compute by @daniel-dodd in #120
• Prevent f64 default by @thomaspinder in #129
• Update w/ Dan comments by @thomaspinder in #130
• Cleanup reqs by @thomaspinder in #131
• Update docs by @thomaspinder in #137
• Improve readability and add comments. by @daniel-dodd in #138
• Move params to the first slot of each function, class, etc. by @daniel-dodd in #139
• V0.5 update by @thomaspinder in #123

Full Changelog: v0.4.13...v0.5

v0.4.13 - Bump docs and dependencies

What's Changed

• Refactor docs as markdown documents by @thomaspinder in #107
• Refactor JaxTyping to be compatible with v0.0.2
• Pin pypa-publish workflow

Full Changelog: v0.4.12...v0.4.13

v0.4.12- Bug fix on bijections

This minor release resolves the issue surrounding Distrax transformations.

v0.4.11 - Parameter seeding

A PRNGKey can now be passed to the initialisation function for reproducible parameter initialisation when parameters are stochastic e.g., RFFs. Further, the return argument of initialise and fit is a dataclass than bundles up the constituent quantities.

v0.4.10

Minor change that fixes a bug in conjugate regression models where the marginal log-likelihood is evaluated on a single datapoint.

v0.4.9 - JOSS Paper & Zenodo

Creating a release for Zenodo archiving and JOSS publication.

v0.4.8 - Titsias Bound

Added implementation and notebook for Titsias 2009.

v0.4.6 - Sparse GPs

In this release the following approaches have been added

• Sparse variational GPs
• Graph kernels
• Variational family abstraction
• A more flexible likelihood representation
• Docs on barycentres of GPs, deep kernels, graph kernels and sparse regression

Contributors - @thomaspinder @daniel-dodd

V0.4 - Parameter refactoring

Major changes

• Parameters are now refactored to impose a greater degree of order and structure.
• Multiple dispatch has been removed for easier extendability

Minor changes

• Optax integration provided
• Matern and Polynomial kernels added
• Constant mean function added
• Global parameter updates
• Significantly slimmer parameter handling code

V0.3 - Pure Jax refactoring

Major changes

• GPJax is now written entirely in Jax with the ObJax dependency removed.
• All notebooks and unit tests have been updated to reflect this new structure.

Minor changes

• ARD kernel structures are now easier to define
• Priors can be placed explicitly on a GP's hyperparameters.