Compared to v2.11.0-pre1:

• Slight adjustments to console output for MultiFits.
• Fixed crashes when trying to plot after a failed fit.
• Adjustments to plot colors for plots with many fits.
• Revised pull plots.
• Enable specifying confidence levels for contours as an alternative to sigma values.

Compared to v2.10.0:

• Dropped Python 3.8, added Python 3.13.
• Changed the model function code to consistently use function signatures instead of bytecode. This makes it possible to override a function's signature for use with kafe2 by simply setting the function's __signature__ property.

Compared to v2.10.0:

• Dropped Python 3.8, added Python 3.13.
• Changed the model function code to consistently use function signatures instead of bytecode. This makes it possible to override a function's signature for use with kafe2 by simply setting the function's __signature__ property.

Features

• Added QR decomposition for numerical stability (0e27fb4), to use the old but faster Cholesky decomposition use the "fast" flag to the cost function (784e0c0)
• Added warnings when starting values/errors are undefined, as well as for very large and small values bad for numerical stability (56e9a84 and fcd1925)
• At least two significant digits for CL annotations in the Contour Profiler are shown (265685a)
• Added initial documentation of the Contour Profiler (2a4d717 and d3cfb14)

Fixes

• Removed the dependency on scipy.misc.derivative which is being removed with SciPy v1.12.0. Instead, numdifftools (which is already a kafe2 dependency anyway) is used (b221b57)
• Fixed example 004 data generation (7e57077)
• Fixed a crash when plotting the error band when at least one parameter is fixed (e3d26cb)

Full Changelog: v2.9.0...v2.10.0

Fixed a crash when plotting the error band when at least one parameter is fixed.

Pre-release for v2.10.0:

• Removed the dependency on scipy.misc.derivative which is being removed with SciPy v1.12.0. Instead, numdifftools (which is already a kafe2 dependency anyway) is used (b221b57)
• At least two significant digits for CL annotations in the Contour Profiler are shown (265685a)
• Fixed example 004 data generation (7e57077)
• Added initial documentation of the Contour Profiler (2a4d717 and d3cfb14)

Full Changelog: v2.9.0...v2.10.0-pre1

Compared to v2.8.3:

• Implemented the ability to do pull plots in the same way you can do ratio and residual plots.
• Extended the wrapper interface to allow the definition of pull, ratio, and residual plots.
• Added a new example under examples/009_histogram_fit/03_SplusBfit.py.

• Fixed #241 (Different behavior depending on parameter names in model)

Fixed a problem with fix parameters in Iminuit which was described in #240

Relative to v2.8.0:

• Matplotlib 3.8 compatibility, minimum Matplotlib version is 3.4
• Added a Makefile
• Moved packaging configuration to pyproject.toml. With this release, Python 3.6 is no longer supported
• Added linting for the project and reformatted many files

Compared to v2.7.0:

• New wrapper functions for CustomFit, HistFit, IndexedFit, and UnbinnedFit.
• All wrapper functions now accept a model function as its first argument and the data afterwards. The interface in v2.7.0 was accidentally built to the wrong specification, and thus had to be remade. The intent is to use the exact same order of arguments as e.g. SciPy curve_fit.
• The interface for creating or plotting profiles has been extended. You can now specify explicit upper or lower bounds as well as confidence levels for the intervals. See examples/003_profiling/04_one_sided_limit.py for an example.
• HistFit/hist_fit can now accept NumPy histograms as data.
• Revised the documentation, particularly the theory chapter.
• Fixed incorrect warning when creating a HistFit.

Compared to v2.8.0-pre1:

• Ability to fix parameters when using wrapper functions.
• Ability to construct HistContainer from only bin edges.
• Better logic for error band and density label defaults.
• Ability to scale fonts in plots.