v2.0.0

Overhaul of kdescent:

• Added a KPretrainer that can be written to and loaded from disk
• KCalc.compare_*_counts() now only needs to compute model evaluations, while the KPretrainer handles data evaluations

v1.5.0

Bug fix: propagate randkey consistently across MPI ranks in run_adam()

Full Changelog: v1.4.0...v1.5.0

v1.4.0

• New feature: compute statistical errors in kdescent
• Add KCalc.reduced_chisq_loss() convenience method
• Remove dependence on deprecated jaxopt library

Full Changelog: v1.3.0...v1.4.0

v1.3.0

• Bug fixes with parallelization and MPI sub-communicator handling

Full Changelog: v1.2.0...v1.3.0

v1.2.0

Further uniformized fitting functions:

• Added missing thin and progress keyword arguments
• All fitting functions now return all params AND losses (thinned, optionally)

v1.1.1

JOSS Publication Release

• Added contributing guidelines to the docs

v1.1.0

Add thin and progress keyword arguments to fitting functions. Default behavior is unchanged (thin=1 and progress=True).

v1.0.0

Initial PyPI release