v0.1.6

Breaking Changes

• acquisition typo fixed, module and class names have been renamed accordingly.
• API changes for BaseOptimizer: n_parallel has been renamed to n_vmap to avoid confusion with n_pmap.

Other changes

• Support for pmap for GridSearch. Also, it technically works with BayesianOptimizer, but the parallel process is independent for each worker (no shared GP kernel). This will be fixed in a future version.
• New notebooks!
• Acquisition is more random: this is to avoid sampling points that are all in the same vicinity. Instead, now we pick i samples out of the top N ranked points.
• GP outputs are normalized at training time.
• Switch from Unittest to PyTest.

Full Changelog: v0.1.5...v0.1.6

v0.1.5

Fix inconsistent English/American spelling in API.

We are now full American 🦅 🇺🇸

v0.1.4

Breaking API changes

• Switched to American Spelling across the board. Function and class names have changed accordingly.
• grid_search.py module has been renamed grid.py.
• bayes.py has been renamed bayesian.py.
• Removed RandomSearch and instead added random_search argument to GridSearch class

Other changes

• Removed pmap functionality as it is currently gigabroken, as you can't use pmap within a scan/fori loop. I will aim to shard the domain instead.

v0.1.3

• Downgraded JAX to 0.4.38.
• Added JAX-tqdm as a dependency.
• Added progress bars to lax inner loops.
• Added Matern kernel and made Matern52 default for BayesOptimisation.
• New logo ✨

Upgrade to v0.1.2a

hyperoptax 0.1.2a - First release

First working version

• Implemented Bayesian Optimisation
• Changed the API for the BaseOptimiser methods
• Various QoL changes

v0.1.1

Initial release

v0.1.0

Initial Release