JuBiotech/calibr8: v6.2.0

This release is a major, but backwards-compatible refactoring that makes calibr8 models more compatible with respect to independent variable dimensionality, and all kinds of continuous or discrete dependent variable distributions.

Calibration models should now inherit from ContinuousUnivariateModel or ContinuousMultivariateModel and a "noise model" that inherits from DistributionMixin. Further reading: https://calibr8.readthedocs.io/en/latest/calibr8_inheritance.html

For example:

class MyModel(calibr8.ContinuousUnivariateModel, calibr8.NormalNoise):
    ...

• Class inheritance was changed & noise model mixins were added, enabling more generalization (also see https://github.com/JuBiotech/calibr8/pull/12)
• On assignment of CalibrationModel.theta_fitted the length of the vector is validated (see https://github.com/JuBiotech/calibr8/pull/17).
• calibr8.HAS_PYMC and from calibr8.utils import pm can be used to condition on the installation of PyMC versions 3 or 4.
• New property CalibrationModel.ndim was added.
• NumericPosterior was replaced by UnivariateInferenceResult and a corresponding hierarchy of more generalized InferenceResult types.