Bring release up to latest stable version

Merge pull request #34 from bb515/develop

Develop

Make explicit the equivalence between discrete markov chains and continuous time numerical solvers

Rerelease of 0.1.6 old that had a bug in VE Markov chains. This changes the api a bit, especially for SDEs, but things are more general and explicit and you can now use a cosine beta schedule (or any beta schedule) too.

0.1.5

The "Bayes update". Added new inverse guidance methods, approximately sample from approximate(p(x|y)) given diffusion model p(x) and data y.

0.1.4

Refactor sde class to be more pythonic (flat).
Added bayesian inversion methods for linear inverse problems.
Added linting, refactored, then removed again since linting was not compatible with python 3.7 or 3.8.

0.1.3

Change sde class structure and add flake8, ruff and mypy

0.1.2

New linting, remove requirements.txt in favour of extras in setup.py, improve CI workflow, change linting so that tab is 2 spaces rather than 4, refactor so that no longer directly depend on orbax or optax, and make clear those dependencies that only occur within the examples.

0.1.1

Markov chains!

0.1.0

Update to 0.1.0. This update includes

• a planned deprecation of meth:utils.update_step
• jit multiple training steps together
• full support for pmap across devices (GPU/TPU) of the same model
• refactoring to make library simpler, flatter
• bug fixes and minor improvements

0.0.9

Made pmap optional