Version 0.5.0

This release adds ABC methods, new gradient-based optimisers and likelihood classes, and improvements to transformations. Testing for Python 3.5 and 3.6 have been dropped in favour of 3.10 and 3.11, so that the minimum supported version is now Python 3.7.

For further changes please consult the CHANGELOG.

Version 0.4.0

This new release adds pints.Transformation and associated subclasses, which provide methods to transform between different representations of a parameter space; for example from a "model space" (p) where parameters have units and some physical counterpart to a "search space" (e.g. q=log(p)) where parameters are non-dimensionalised and less-recognisable to the modeller. Parameter transformation can be useful in many situations, for example transforming from a constrained parameter space to an unconstrained search space using RectangularBoundariesTransformation leads to crucial performance improvements for many methods.

For other changes please consult the CHANGELOG

Version 0.3.0

This is the first release in a while for PINTS, and while it's still not a 1.0.0 version, the code has been quite stable, and growth has mainly been through the addition of new features! These include:

• New MCMC sampling methods, including slice sampling methods, relativistic and monomial gamma HMC, and an overhaul of the "adaptive covariance MCMC" method, which now supports several variants.
• New optimisers (added for insight, not performance), including Nelder-Mead, a gradient descent method and a bare-bones implementation of CMA-ES
• Several new error measures, log likelihoods, and log priors.
• New toy problems, such as the eight-schools problem, the german credit problem, stochastic degradation, and the simple harmonic oscillator
• Noise generating methods, which can be combined with the new autoregressive likelihoods to investigate the consequences of choosing different noise models
• New diagnostic plots
• Support for arbitrarily shaped boundaries
• An MCMCSummary object, and a fix to the rhat method
• Various bugfixes, improvements to documentation, and new example notebooks

Finally, PINTS is now on PyPI, so that you'll be able to install it with pip install pints without first downloading the repository

Version 0.2.2

Many new methods, and some changes to API, including addition of methods/models using 1st order sensitivities, and renaming of MCMCSampling to MCMCController and Optimisation to OptimisationController

Version 0.1.1

This second pre-release contains a number of bugfixes and performance improvements. A notable change to the API is that the LogLikelihood class has been removed.
Please note that Pints in still under active development, and so the API may change in the future.

Version 0.1.0

This is the first official version of Pints. Please note that Pints in still under active development, and so the API may change in the future.