Iminuit

-------------------

New features
~~~~~~~~~~~~
- Builtin cost functions now report the number of data points with the attribute
``Cost.ndata``
- New attribute ``Minuit.ndof`` returns the degrees of freedom if the cost function
reports it or NaN
- New attribute ``FMin.reduced_chi2`` to report the reduced chi2 of the fit; returns
NaN if the reduced chi2 cannot be computed for the cost function, in case of unbinned
maximum-likelihood or when the attribute ``Cost.ndata`` is missing

----------------------

New features
~~~~~~~~~~~~
- ``util.merge_signatures`` added based on ``merge_user_func`` from ``probfit``,
by mbaak
- ``util.make_with_signature`` added to create new functions with renamed arguments
- ``util.BasicView.to_dict`` added, by watsonjj
- ``util.BasicView`` and ``util.Matrix`` now supports element selection with sequences
like ``numpy.ndarray``
- ``util.propagate`` to error propagate covariance matrices from one vector space to
another (Jacobi matrix is computed numerically)

Fixes
~~~~~
- ``util.BasicView`` now supports slices of the form ``a[:len(a)]`` or ``a[:M]`` with
``M > len(a)`` like other Python containers
- ``util.Matrix`` now returns a square matrix when it is used with a slice or item
selection
- Missing comma in BibTeX entry shown in CITATION.rst, by Ludwig Neste

Other
~~~~~
- ``util.describe`` returns list instead of tuple

Documentation
~~~~~~~~~~~~~
- Better docstring for ``util.FMin``
- New tutorial on how to do simultaneous fits / adding likelihoods, by watsonjj
- New tutorial on how to use builtin cost function
- New tutorial about how to draw error bands around fitted curves

-------------------------

New features
~~~~~~~~~~~~
- ``minimize``
- Keyword ``method`` now accepts "migrad" and "simplex"
- Keyword ``option`` now supports keyword "stra" to set ``Minuit.strategy``
- ``OptimizeResult.message`` now states if errors are not reliable
- ``Minuit`` now supports functions wrapped with ``functools.partial``, by jnsdrtlf

Other
~~~~~
- Upgrade Minuit2 C++ code in ROOT to latest version with following improvements

- improvement of seed when using an analytical gradient
- fix of last minimum state added twice to vector of minimum states in some cases
(no impact for iminuit users, but saves a bit of memory)

- Documentation improvements
- Updated tutorial about automatic differentiation, added comparison of ``numba.njit``
and ``jax.jit``

------------------------

New features
~~~~~~~~~~~~
- ``cost.BinnedNLL`` and ``cost.ExtendedBinnedNLL`` now support
weighted binned data

Bug-fixes
~~~~~~~~~
- ``FMin.edm_goal`` now remains unchanged if ``Minuit.hesse`` is run after
``Minuit.migrad``

Other
~~~~~
- Update to cibuildwheels-1.8.0 and workflow simplification, by henryiii

-------------------------

Minor improvements
~~~~~~~~~~~~~~~~~~
- ``Minuit.profile``, ``Minuit.mnprofile``, ``Minuit.contour``, ``Minuit.draw_profile``,
``Minuit.draw_mnprofile``, and ``Minuit.draw_contour`` can now be called with
``subtract_min=True`` even if ``Minuit.fmin`` is None
- ``__version__`` now also displays the ROOT version of the C++ Minuit2 library
- Support for adding constant numbers to cost functions, this allows you to write
``sum(cost1, cost2, ...)`` and may be useful to subtract a constant bias from the
cost

Other
~~~~~
- Documentation improvements

- Further transition to numpydoc
- Clarified that iminuit is based on ROOT code
- List full iminuit version including ROOT version in docs

- Added type hints to many interfaces (incomplete)
- Renamed ``_minuit`` to ``minuit``, making the module public
- Renamed ``_minimize`` to ``minimize``, making the module public
- pydocstyle added to pre-commit checks

-------------------------

New features
~~~~~~~~~~~~
- Cost functions in ``cost`` are now additive, creating a new cost function with the
union of parameters that returns the sum of the results of the individual cost functions
- ``cost.NormalConstraint`` was added as a means to add soft constraints on a
parameter, can also be used to set up a covariance matrix between several parameters

Other
~~~~~
- Documentation improvements, started transition to numpydoc