Numdifftools

===========================
Fabian Joswig (5):
* ci: execute test action only on push to master and on pull requests.
* ci: test requirements added to ci workflow.
* ci: first version of github actions ci added.
* fix: import from from scipy.ndimage.filters replaced by from scipy.ndimage
* fix: np.info(float).machar.tiny replaced by np.info(float).tiny

Jonas Eschle (6):
* Drop Python 3.6
* Remove Python 2.7, 3.6 from appveyor CI
* Update .travis.yml
* Update setup.cfg
* Update .travis.yml
* Update to Python310

Per A Brodtkorb (19):
* Commented out deprecated pep8ignore and pep8maxlinelength in setup.cfg
* Fixed issue 59: numpy deprecation warning on machar.tiny
* Deleted obsolete travis_install.sh
* Replaced deprecated `np.MachAr().eps` (NumPy 1.22) with np.finfo(float).eps in test_multicomplex.py

* Added requirements.tests.txt
* Updated .github/workflows/test.yml to use requirements.tests.txt
* Removed obsolete .travis.yml and appveyor.yml.
* Github-actions are now used instead.
* Replaced appveyor badge and travis badge with github-actions badge in README.rst, info.py and index.rst
* Removed python 2.7 from classifiers in setup.cfg
* Updated .travis.yml
* Fixed doctest so they don't crash on travis: Replaced " doctest + SKIP" with " doctest: +SKIP" in docstrings.
* Updated download badge in README.rst and info.py
* Updated test_img in README.rst
* Updated tests_img path for travis.
* Added " doctest + SKIP" to doctest string in info.py
* Replaced "version|" with "release|" in docs/index.rst
* Added matplotlib to requirements.txt Removed failing python 3.8 from appveyor.yml

Per A. Brodtkorb (4):
* Merge pull request 65 from fjosw/feat/github_actions_ci
* Merge pull request 66 from jonas-eschle/patch-1
* Merge pull request 60 from peendebak/performance/percentile
* Merge pull request 63 from fjosw/feat/numpy_deprecation

Pieter Eendebak (2):
* workaround for known issue with np.nanpercentile
* improve performance by combining percentile calculations

==========================

Per A Brodtkorb (109):
* Replaced python 3.5 with 3.9 in .travis.yml
* Removed python 3.5 from appveyor.yml
* Added bibtex_bibfiles = ... to docs/conf.py
* Fixed doctest failures in
- docs/src/numerical/derivest.rst
- docs/tutorials/getting_started.rst
- numdifftools.core.py
- numdifftools.limits.py
- numdifftools.nd_algopy.py
- numdifftools.nd_scipy.py
- numdifftools.nd_statsmodels.py
* Insulated import of click in a if __name__ =='__main__' clause.
* Added activate to appveyor.yml
* Added https://mathworld.wolfram.com/WynnsEpsilonMethod.html reference for the Epsilon algorithm in extrapolation.py.
* Disabled the restriction that n must be one in LogJacobianRule
* Added complex_even and central_even methods to the JacobianDifferenceFunctions
* Updated documentation of Derivative in core.py
* Updated documentation of Richardson.
* Removed obsolete tests from test_nd_scipy.py
* Fixed a bug in TestJacobian.test_scalar_to_vector in test_nd_scipy.py for method="complex'
* Refactored code from core.py to finite_difference.py
* Added LogJacobianRule, LogHessdiagRule, LogHessianRule to finite_difference.py
* Fixed a bug in Richardson._estimate_error: Complex rule resulted wrongly in complex error values.
* Added netlib.org/quadpack reference.
* Updated Dea to conform with Quadpack
* Replaced reference to Brezinski with refs to Quadpack.
* Reduced cyclomatic complexity in Dea in extrapolation.py
* Removed commented out code in profile_numdifftools.py
* Updated pycodestyle ignore section in setup.cfg
* Removed commented out code in run_benchmark.py Made get_nominal_step continous as function of x
* Made datetime call python 2.7 compatible in run_benchmark.py
* Simplified the Derivative._step_generator function in core.py.
* Made plots generated from run_benchmark.py prettier.
* step_ratio in the step generators by default 2 for n=1 and 1.6 otherwise in step_generators.py
* Fixed failing doctests in core.py and nd_statsmodels.py
* Added doctests to setup.cfg.
* Reordered imports in test_example_functions.py
* Fixed .travis.yml so that he file paths in coverage.xml is discoverable
under the sonar.sources folder. The problem is that SonarQube is
analysing the checked-out source code (in src/numdifftools) but the
actual unit tests and coverage.py is run against the installed code (in
build/lib/numdifftools). Thus the absolute files paths to the installed
* Removed commented code from test_numdifftools.py
* Run only coverage xml when python version is 3.7
* Updated .travis.yml Removed commented out code from extrapolation.py and nd_statsmodels.py
* Finalized the moved of XXXDifferencdFunctions from core.py to finite_difference.py
* Added missing docstring for default_scale function in step_generators.py.
* Removed unused import of itertools in _find_default_scale.py.
* Changed default scale from 1.35 to 1.06 for complex/multicomplex methods when n=1
* Added richardson_demo to extrapolation.py Simplified EpsAlg class in extrapolation.py
* Corrected a small error for dea3: Now dea3 and Dea(limexp=3) gives the same result!
* Added python 3.8 to appveyor.yml Added python 3.9 to setup.cfg
* Fixed reference to how-to/index
* Added doctest configuration to docs.conf.py
* Fixes issue 50 by adding function value f(x) to the info.f_value.
* Updated README.rst
* Added UnusedVariable here and there.
* Silence warnings in Hessian by adding __init__ that set the correct order given the method.
* Updated the Richardson._r_matrix method to generate complex matrix when step_ratio is complex.
* Fixed profile_hessian in profile_numdifftools.py so it works again.
* Added with np.errstate(all='ignore') to test_derivative_on_sinh and test_scalar_to_vector in test_nd_algopy.py to silence warnings.
* Changed citation style to alpha.
* Replaced bibliography.rst with refs1.bib and zreferences.rst
* Removed badges for latex
* Changed sonar addon token
* Added CC_TEST_REPORTER_ID
* Fixed a typo in docs/numdifftools.rst
* Added docs/make.bat
* Removed python 2.7 from .travis.yml
* Moved test_requires from setup.cfg to setup.py
* Added Latex to setup.py
* Changed default radius to 0.0059 which appears to cause less failures in Taylor in fornberg.py.
* Updated MANIFEST.in
* Fixes issue 49 : Dimension of Jacobian of vector valued function (length n) with scalar input should be n X 1
* Updated build_package.py
* Attempt to silence divide by zero and invalid warnings.
* Fix issue52: Gradient tries to apply squeeze to the output tuple containing both the result and the full_output object.
* Made docstring a rawdocstring since it contains slashes.
* Added " pylint: disable=unused-argument" in appropriate places.
* API change: replaced "python setup.py doctests" with "python setup.py doctest"
* Removed unused imports
* Fixed a bug in test_low_order_derivative_on_example_functions: Same variable (i) was used both in the outer and inner loop.
* Updated badge for pypi and documentation of fornberg.py
* Fixed failing tests.
* Updated docs + added a test
* Added "python -m pip install --upgrade pytest" to appveyor.yml due to a package conflict on python2.7 32 bit
* Added - "python -m pip install --upgrade setuptools" in appveyor.yml to avoid build error.
* Try "python setup.py bdist_wheel" and "pip install numdifftools --find-links=dist" in appveyor.yml
* Put qoutes on "python -m pip install --upgrade pip" in appveyor.yml
* Changed "python setup.py install" to
- python setup.py bdist_wheel"
- pip install numdifftools --find-links=dist
* Added "pip install --upgrade pip" to appveyor.yml
* Updated the detailed package documentation.
* Added missing pytest-pep8 to install
* Updated badge + appveyor.yml
* ongoing work to harmonize the the output from approx_fprime and approx_fprime_cs
* Added Taylor class to nd_algopy.py Fixed a bug in _get_best_taylor_coefficient in fornberg.py
* Updated references Added test_mod_c function to test_multicomplex.py
* Fixed a typo.
* Removed --strict-markers
* Fixed issue 39 TypeError: unsupported operand type(s) for /: 'float' and 'Bicomplex'
* Fixed a typo in the documentation. Closing issue 51
* Added separate test for nd_scipy.
* added skip on tests if LineProfiler is not installed.
* Removed obsolete centered argument from call to approx_hess1 + pep8
* Move Jacobian._increment method to _JacobianDifferenceFunctions
* step_nom was unused in CStepGenerator.__init__ Added pytest.markers.slow in to setup.cfg
* Made two tests more forgiving in order to avoid failure on travis.
* Renamed nominal_step and base_step to get_nominal_step and get_base_step, respectively.
* Removed obsolete import of example from hypothesis
* Updated testing
* Updated coverage call: coverage run -m py.test src/numdifftools/tests
* Delete obsolete conftest.py

===========================

Robert Parini (1):
* Fix issue 43: numpy future warning

===========================

Andrew Nelson (1):
* MAINT: special.factorial instead of misc.factorial

Dougal J. Sutherland (1):
* include LICENSE.txt in distributions

Per A Brodtkorb (140):
* Adjusted runtime for hypothesis tests to avoid failure and fixed pep8 failures.
* Fixed a bug in setup.cfg
* Replaced valarray function with numpy.full in step_generators.py
* Added try except on import of algopy
* Updated the badges used in the README.rst
* Replaced numpy.testing.Tester with pytest.
* Removed dependence on pyscaffold.
* Simplified setup.py and setup.cfg
* Updated .travis.yml configuration.
* Reorganized the documentation.
* Ongoing work to simplify the classes.
* Replaced unittest with pytest.
* Added finite_difference.py
* replaced , with .
* Reverted to coverage=4.3.4
* New attempt
* Fixed conflicting import
* Missing import of EPS
* Added missing FD_RULES = {}
* Removed pinned coverage, removed dependence on pyscaffold
* Updated .travis.yml and .appveyor.yml
* Replaced conda channel omnia with conda-forge
* Removed commented out code. Set pyqt=5 in appveyor.yml
* Updated codeclimate checks
* Dropped support for python 3.3 and 3.4. Added support for python 3.6, 3.7
* Simplified code.
* Pinned IPython==5.0 in order to make the testserver not crash.
* Added line_profiler to appveyor.yml
* Removed line_profiler from requirements.txt
* Fix issue 37: Unable to install on Python 2.7
* Added method='backward' to nd_statsmodels.py
* Skip test_profile_numdifftools_profile_hessian and TestDoProfile
* Added missing import of warnings
* Added tests for the scripts from profile_numdifftools.py, _find_default_scale.py and run_benchmark.py.
* Added reason to unittest.skipIf
* Added line_profiler to requirements.
* misssing import of warnings fixed.
* Renamed test so it comes last, because I suspect this test mess up the coverage stats.
* Reordered the tests.
* Added more tests.
* Cleaned up _find_default_scale.py
* Removed link to depsy
* Reverted: install of cython and pip install setuptools
* Disabled sonar-scanner -X for python 3.5 because it crashes.
* Reverted [options.packages.find] to exclude tests again
* Added cython and reverted to pip install setuptools
* Updated sphinx to 1.6.7
* Try to install setuptools with conda instead.
* Added hypothesis and pytest to requirements.readthedocs.txt
* Set version of setuptools==37.0
* Added algopy, statsmodels and numpy to requirements.readthedocs.txt
* Restricted sphinx in the hope that the docs will be generated.
* Removed exclusion of tests/ directory from test coverage.
* Added dependencies into setup.cfg
* Readded six as dependency
* Refactored and removed commented out code.
* Fixed a bug in the docstring example: Made sure the shape passed on to zeros is an integer.
* Fixed c_abs so it works with algopy on python 3.6.
* Fixed flaky test and made it more robust.
* Fixed bug in .travis.yml
* Refactored the taylor function into the Taylor class in order to simplify the code.
* Fixed issue 35 and added tests
* Attempt to simplify complexity
* Made doctests more robust
* Updated project path
* Changed install of algopy
* Fixed small bugs
* Updated docstrings
* Changed Example and Reference to Examples and References in docstrings to comply with numpydoc-style.
* Renamed CHANGES.rst to CHANGELOG.rst
* Renamed source path
* Hack due to a bug in algopy or changed behaviour.
* Small fix.
* Try to reduce complexity
* Reduced cognitive complexity of min_num_steps
* Simplified code in Jacobian
* Merge branch 'master' of https://github.com/pbrod/numdifftools
* Fixed issue 34 Licence clarification.
* Locked coverage=4.3.4 due to a bug in coverage that cause code-climate test-reporter to fail.
* Added script for finding default scale
* updated from sonarcube to sonarcloud
* Made sure shape is an integer.
* Refactored make_step_generator into a step property
* Issue warning message to the user when setting the order to something different than 1 or 2 in Hessian.
* Updated example in Gradient.
* Reverted --timid option to coverage because it took too long time to run.
* Reverted --pep8 option
* pep8 + added --timid in .travis.yml coverage run in order to to increase missed coverage.
* Refactored taylor to reduce complexity
* No support for python 3.3. Added python 3.6
* Fixed a small bug and updated test.
* Removed unneccasarry perenthesis. Reduced the complexity of do_profile
* Made python3 compatible
* Removed assert False
* Made unittests more forgiving.
* updated doctest in nd_scipy.py and profiletools.py install line_profiler on travis
* Made python 3 compatible
* Updated tests
* Added test_profiletools.py and capture_stdout_and_stderr in testing.py
* Optimized numdifftools.core.py for speed: fd_rules are now only computed once.
* Only keeping html docs in the distribution.
* Added doctest and updated .pylintrc and requirements.txt
* Reduced time footprint on tests in the hope that it will pass on Travis CI.
* Prefer static methods over instance methods
* Better memory handling: This fixes issue 27
* Added statsmodels to requirements.txt
* Added nd_statsmodels.py
* Simplified input
* Merge branch 'master' of https://github.com/pbrod/numdifftools
* Updated link to the documentation.

Robert Parini (4):
* Avoid RuntimeWarning in _get_logn
* Allow fd_derivative to take complex valued functions

solarjoe (1):
* doc: added nd.directionaldiff example

============================

Per A Brodtkorb (1):
* Updated the author email address in order for twine to be able to upload to pypi.

============================

Per A Brodtkorb (1):
* Updated setup.py in a attempt to get upload to pypi working again.