Tags

- Updating sampling for distributions, updating sampling for incomplete grids
- Optimizing performance tuning
- All instance of nchains were changed to njobs, including nsamples_per_chain -> nsamples_per_job
- There is no recommended value any more for the number of samples per chain. It is advised to start with a small number and increase as needed to get satisfactory diagnostics. When no number is provided by the user, the default will be the minimum number set in rcparams.

- fixing bugs for incomplete grids + parameters following a statistical distribution
- fixing bugs for GPU use
- 1st version "officially" supported for GPU backend (mgris using PyMC using blackjax using jax using XLA to talk to the GPU)

Major upgrade with potential breaking changes
- core function for interpolation/integration within grid completely rewritten
- more functions can now use jax as an accelerated numpy (when jax is available, it is not mandatory)
- first version with (some) GPU capabilities (still in testing phase) for SMC and NUTS.

Main change concerns the statistical distribution of components (power-law, normal etc...) that now use a new, much narrower, scaling prior to match the grid values to the observations. Plots have also been improved.

Fixed a bug for linear interpolation of post-processed parameters with incomplete grids.

This is the first release of MULTIGRIS, tested with PyMC versions until 5.

Release candidate 6 for 1.0.0

Release candidate 5 for 1.0.0

Release candidate 4 for 1.0.0

Release candidate 3 for 1.0.0

Release candidate 2 with new priors for scaling

Version 1.0.0, release candidate 1