This alpha release is an unstable working release for the project **reversible-damage**.
What's Changed
* Implementing `Evaluator` classes for safe dispatched model evaluation by flo-schu in https://github.com/flo-schu/pymob/pull/5
* :boom: **This breaks `simulation.run` and `simulation.evaluate`**. These methods are not used any longer. Instead Evaluators are dispatched from the simulation `simulation.dispatch(theta={"a":1, "b":2}, **kwargs)`, where a and b are parameters of the model and kwargs are keyword arguments processed by the model solver.
* :boom: **Requires now that simulation results are returned as dictionaries so that they can be safely mapped to `xarray.Datasets`**. This implementation detail may change in future releases.
* Implementing full Byesian inference with `numpyro` + `jax` :rocket: 8 and a corresponding solver based on `diffrax` in the test case study.
* Automatic parsing of error models and priors into numpyro model syntax 13 :tada:
* Harmonizes API and tests `pyabc` and `numpyro` inference on *Lotka-Volterra* models
* Documents numypro and adds `sphinx` documentation for publishing on readthedocs
* Adds the `solvetools` module which provides Solver classes and a `mappar` function, to derive ODE model arguments from the signature and extract them from a dictionary of global parameters.
**Full Changelog**: https://github.com/flo-schu/pymob/compare/0.2.5...0.3.0a3