Solves automatic numerical differentiation problems in one or more variables.
Project description
numdifftools
The numdifftools library is a suite of tools written in _Python to solve automatic numerical differentiation problems in one or more variables. Finite differences are used in an adaptive manner, coupled with a Richardson extrapolation methodology to provide a maximally accurate result. The user can configure many options like; changing the order of the method or the extrapolation, even allowing the user to specify whether complexstep, central, forward or backward differences are used.
The methods provided are:
Derivative: Compute the derivatives of order 1 through 10 on any scalar function.
directionaldiff: Compute directional derivative of a function of n variables
Gradient: Compute the gradient vector of a scalar function of one or more variables.
Jacobian: Compute the Jacobian matrix of a vector valued function of one or more variables.
Hessian: Compute the Hessian matrix of all 2nd partial derivatives of a scalar function of one or more variables.
Hessdiag: Compute only the diagonal elements of the Hessian matrix
All of these methods also produce error estimates on the result.
Numdifftools also provide an easy to use interface to derivatives calculated with in _AlgoPy. Algopy stands for Algorithmic Differentiation in Python. The purpose of AlgoPy is the evaluation of higherorder derivatives in the forward and reverse mode of Algorithmic Differentiation (AD) of functions that are implemented as Python programs.
Getting Started
Visualize high order derivatives of the tanh function
>>> import numpy as np >>> import numdifftools as nd >>> import matplotlib.pyplot as plt >>> x = np.linspace(2, 2, 100) >>> for i in range(10): ... df = nd.Derivative(np.tanh, n=i) ... y = df(x) ... h = plt.plot(x, y/np.abs(y).max())>>> plt.show() # doctest: +SKIP
Compute 1’st and 2’nd derivative of exp(x), at x == 1:
>>> fd = nd.Derivative(np.exp) # 1'st derivative >>> fdd = nd.Derivative(np.exp, n=2) # 2'nd derivative >>> np.allclose(fd(1), 2.7182818284590424) True >>> np.allclose(fdd(1), 2.7182818284590424) True
Nonlinear least squares:
>>> xdata = np.reshape(np.arange(0,1,0.1),(1,1)) >>> ydata = 1+2*np.exp(0.75*xdata) >>> fun = lambda c: (c[0]+c[1]*np.exp(c[2]*xdata)  ydata)**2 >>> Jfun = nd.Jacobian(fun) >>> np.allclose(np.abs(Jfun([1,2,0.75])), 0) # should be numerically zero True
Compute gradient of sum(x**2):
>>> fun = lambda x: np.sum(x**2) >>> dfun = nd.Gradient(fun) >>> np.allclose(dfun([1,2,3]), [ 2., 4., 6.]) True
Compute the same with the easy to use interface to AlgoPy:
>>> import numdifftools.nd_algopy as nda >>> import numpy as np >>> fd = nda.Derivative(np.exp) # 1'st derivative >>> fdd = nda.Derivative(np.exp, n=2) # 2'nd derivative >>> np.allclose(fd(1), 2.7182818284590424) True >>> np.allclose(fdd(1), 2.7182818284590424) True
Nonlinear least squares:
>>> xdata = np.reshape(np.arange(0,1,0.1),(1,1)) >>> ydata = 1+2*np.exp(0.75*xdata) >>> fun = lambda c: (c[0]+c[1]*np.exp(c[2]*xdata)  ydata)**2 >>> Jfun = nda.Jacobian(fun, method='reverse') >>> np.allclose(np.abs(Jfun([1,2,0.75])), 0) # should be numerically zero True
Compute gradient of sum(x**2):
>>> fun = lambda x: np.sum(x**2) >>> dfun = nda.Gradient(fun) >>> np.allclose(dfun([1,2,3]), [ 2., 4., 6.]) True
See also
scipy.misc.derivative
Documentation and code
Numdifftools works on Python 2.7+ and Python 3.0+.
Official releases available at: http://pypi.python.org/pypi/numdifftools
Official documentation available at: http://numdifftools.readthedocs.io/en/latest/
Bleeding edge: https://github.com/pbrod/numdifftools.
Installation
If you have pip installed, then simply type:
$ pip install numdifftools
to get the lastest stable version. Using pip also has the advantage that all requirements are automatically installed.
Unit tests
To test if the toolbox is working paste the following in an interactive python session:
import numdifftools as nd nd.test('doctestmodules', 'disablewarnings')
Acknowledgement
The numdifftools package for Python was written by Per A. Brodtkorb based on the adaptive numerical differentiation toolbox written in Matlab by John D’Errico [DErrico06].
Later the package was extended with some of the functionality found in the statsmodels.tools.numdiff module written by Josef Perktold [JPerktold14] which is based on [Rid09]. The implementation of bicomplex numbers is based on the matlab implementation described in the project report of [Ver14] which is based on [GLD12]. For completeness the [For98] method for computing the weights and points in general finite difference formulas as well as the [For81] method for cumputing the taylor coefficients of complex analytic function using FFT, was added.
References
Perktold, J (2014), numdiff package http://statsmodels.sourceforge.net/0.6.0/_modules/statsmodels/tools/numdiff.html
Adriaen Verheyleweghen, (2014) “Computation of higherorder derivatives using the multicomplex step method”, Project report, NTNU
Gregory Lantoine, R.P. Russell, and T. Dargent (2012) “Using multicomplex variables for automatic computation of highorder derivatives”, ACM Transactions on Mathematical Software, Vol. 38, No. 3, Article 16, April 2012, 21 pages, http://doi.acm.org/10.1145/2168773.2168774
M.E. LunaElizarraras, M. Shapiro, D.C. Struppa1, A. Vajiac (2012), “Bicomplex Numbers and Their Elementary Functions”, CUBO A Mathematical Journal, Vol. 14, No 2, (6180). June 2012.
Gregory Lantoine (2010), “A methodology for robust optimization of lowthrust trajectories in multibody environments”, Phd thesis, Georgia Institute of Technology
Ridout, M.S. (2009) “Statistical applications of the complexstep method of numerical differentiation”, The American Statistician, 63, 6674
D’Errico, J. R. (2006), “Adaptive Robust Numerical Differentiation”, http://www.mathworks.com/matlabcentral/fileexchange/13490adaptiverobustnumericaldifferentiation
K.L. Lai, J.L. Crassidis, Y. Cheng, J. Kim (2005), “New complex step derivative approximations with application to secondorder kalman filtering”, AIAA Guidance, Navigation and Control Conference, San Francisco, California, August 2005, AIAA20055944.
B. Fornberg (1998) “Calculation of weights_and_points in finite difference formulas”, SIAM Review 40, pp. 685691.
Fornberg, B. (1981). “Numerical Differentiation of Analytic Functions”, ACM Transactions on Mathematical Software (TOMS), 7(4), 512526. http://doi.org/10.1145/355972.355979
Lyness, J. M., Moler, C. B. (1969). “Generalized Romberg Methods for Integrals of Derivatives”, Numerische Mathematik.
Lyness, J. M., Moler, C. B. (1966). “Vandermonde Systems and Numerical Differentiation”, Numerische Mathematik.
NAG Library. NAG Fortran Library Document: D04AAF
Changelog
Version 0.9.41 Nov 10, 2022
 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.
Githubactions are now used instead.
Replaced appveyor badge and travis badge with githubactions 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 jonaseschle/patch1
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
Version 0.9.40 Jun 2, 2021
 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 checkedout 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 howto/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 issue#52: 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=unusedargument” 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 –findlinks=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 –findlinks=dist
Added “pip install –upgrade pip” to appveyor.yml
Updated the detailed package documentation.
Added missing pytestpep8 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 –strictmarkers
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
Version 0.9.39 Jun 10, 2019
 Robert Parini (1):
Fix issue #43: numpy future warning
Version 0.9.38 Jun 10, 2019
 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 condaforge
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 sonarscanner 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 numpydocstyle.
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 codeclimate testreporter 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
Version 0.9.20, Jan 11, 2017
 Per A Brodtkorb (1):
Updated the author email address in order for twine to be able to upload to pypi.
Version 0.9.19, Jan 11, 2017
 Per A Brodtkorb (1):
Updated setup.py in a attempt to get upload to pypi working again.
Version 0.9.18, Jan 11, 2017
 Per A Brodtkorb (38):
Updated setup
Added import statements in help header examples.
Added more rigorous tests using hypothesis.
Forced to use wxagg backend
Moved import of matplotlib.pyplot to main in order to avoid import error on travis.
Added fd_derivative function
Updated references.
Attempt to automate sonarcube analysis
Added testcoverage to sonarqube and codeclimate
Simplified code
Added .pylintrc + pep8
Major change in api: class member variable self.f changed to self.fun
Fixes issue #25 (Jacobian broken since 0.9.15)
Version 0.9.17, Sep 8, 2016
 Andrew Fowlie (1):
Fix ReadTheDocs link as mentioned in #21
 Per A Brodtkorb (79):
Added test for MinMaxStepgenerator
Removed obsolete docs from core.py
Updated appveyor.yml
Fixed sign in inverse matrix
Simplified code
Added appveyor badge + synchronised info.py with README.rst.
Removed plot in help header
Added Programming Language :: Python :: 3.5
Simplified code
Renamed bicomplex to Bicomplex
Simplified example_functions.py
 Moved MinStepGenerator, MaxStepGeneretor and MinMaxStepGenerator to step_generators.py
Unified the step generators
Moved step_generator tests to test_step_generators.py
Major simplification of step_generators.py
Removed duplicated code + pep8
Moved fornberg_weights to fornberg.py + added taylor and derivative
Fixed print statement
Replace xrange with range
Added examples + made computation more robust.
Made ‘backward’ and alias for ‘reverse’ in nd_algopy.py
Expanded the tests + added test_docstrings to testing.py
Replace string interpolation with format()
Removed obsolete parameter
Smaller start radius for Fornberg method
Simplified “n” and “order” properties
Simplified default_scale
Removed unecessary parenthesis and code.
Fixed a bug in Dea + small refactorings.
Added test for EpsAlg
Avoid mutable default args and prefer static methods over instancemeth.
Refactored to reduce cyclomatic complexity
Changed some instance methods to static methods
Renamed nonpythonic variable names
Turned on xvfb (X Virtual Framebuffer) to imitate a display.
Added extra test for Jacobian
Replace lambda function with a def
Removed unused import
Added test for epsalg
Fixed test_scalar_to_vector
Updated test_docstrings
Version 0.9.15, May 10, 2016
 Cody (2):
Migrated % string formating
Migrated % string formating
 Per A Brodtkorb (28):
Updated README.rst + setup.cfg
Replaced instance methods with static methods +pep8
Merge branch ‘master’ of https://github.com/pbrod/numdifftools
Fixed a bug: replaced missing triple quote
Added depsy badge
added .checkignore for quantificode
Added .codeclimate.yml
Fixed failing tests
Changed instance methods to static methods
Made untyped exception handlers specific
Replaced local function with a static method
Simplified tests
Removed duplicated code Simplified _Derivative._get_function_name
exclude tests from testclimate
Renamed test_functions.py to example_functions.py Added test_example_functions.py
 Per A. Brodtkorb (2):
Merge pull request #17 from pbrod/autofix/wrapped2_to3_fix
Merge pull request #18 from pbrod/autofix/wrapped2_to3_fix0
 pbrod (17):
updated conf.py
added numpydoc>=0.5, sphinx_rtd_theme>=0.1.7 to setup_requires if sphinx
updated setup.py
added requirements.readthedocs.txt
Updated README.rst with info about how to install it using conda in an anaconda package.
updated conda install description
Fixed number of arguments so it does not differs from overridden ‘_default_base_step’ method
Added codecov to .travis.yml
Attempt to remove coverage of testfiles
Added directionaldiff function in order to calculate directional derivatives. Fixes issue #16. Also added supporting tests and examples to the documentation.
Fixed isssue #19 multiple observations mishandled in Jacobian
Moved rosen function into numdifftools.testing.py
updated import of rosen function from numdifftools.testing
Simplified code + pep8 + added TestResidue
Updated readme.rst and replaced string interpolation with format()
Cleaned Dea class + pep8
Updated references for Wynn extrapolation method.
Version 0.9.14, November 10, 2015
 pbrod (53):
Updated documentation of setup.py
Updated README.rst
updated version
Added more documentation
Updated example
Added .landscape.yml updated .coveragerc, .travis.yml
Added coverageall to README.rst.
updated docs/index.rst
Removed unused code and added tests/test_extrapolation.py
updated tests
Added more tests
Readded c_abs c_atan2
Removed dependence on wheel, numpydoc>=0.5 and sphinx_rtd_theme>=0.1.7 (only needed for building documentation)
updated conda path in .travis.yml
added omnia channel to .travis.yml
Added conda_recipe files Filtered out warnings in limits.py
Version 0.9.13, October 30, 2015
 pbrod (21):
Updated README.rst and CHANGES.rst.
updated Limits.
Made it possible to differentiate complex functions and allow zero’th order derivative.
BUG: added missing derivative order, n to Gradient, Hessian, Jacobian.
Made test more robust.
Updated structure in setup according to pyscaffold version 2.4.2.
Updated setup.cfg and deleted duplicate tests folder.
removed unused code.
Added appveyor.yml.
Added required appveyor install scripts
Fixed bug in appveyor.yml.
added wheel to requirements.txt.
updated appveyor.yml.
Removed import matplotlib.
 Justin Lecher (1):
Fix min version for numpy.
 kikocorreoso (1):
fix some prints on run_benchmark.py to make it work with py3
Version 0.9.12, August 28, 2015
pbrod (12):
Updated documentation.
Updated version in conf.py.
Updated CHANGES.rst.
Reimplemented outlier detection and made it more robust.
Added limits.py with tests.
Updated main tests folder.
Moved Richardson and dea3 to extrapolation.py.
Making a new release in order to upload to pypi.
Version 0.9.11, August 27, 2015
 pbrod (2):
Fixed sphinxbuild and updated docs.
Fixed issue #9 Backward differentiation method fails with additional parameters.
Version 0.9.10, August 26, 2015
 pbrod (7):
Fixed sphinxbuild and updated docs.
Added more tests to nd_algopy.
Dropped support for Python 2.6.
Version 0.9.4, August 26, 2015
 pbrod (7):
Fixed sphinxbuild and updated docs.
Version 0.9.3, August 23, 2015
 Paul Kienzle (1):
more useful benchmark plots.
 pbrod (7):
Fixed bugs and updated docs.
Major rewrite of the easy to use interface to Algopy.
Added possibility to calculate n’th order derivative not just for n=1 in nd_algopy.
Added tests to the easy to use interface to algopy.
Version 0.9.2, August 20, 2015
 pbrod (3):
Updated documentation
Added parenthesis to a call to the print function
Made the test less strict in order to pass the tests on Travis for python 2.6 and 3.2.
Version 0.9.1, August 20,2015
 Christoph Deil (1):
Fix Sphinx build
 pbrod (47):
 Total remake of numdifftools with slightly different call syntax.
Can compute derivatives of order up to 1014 depending on function and method used.
Updated documentation and tests accordingly.
Fixed a bug in dea3.
Added StepsGenerator as an replacement for the adaptive option.
Added bicomplex class for testing the complex step second derivative.
Added fornberg_weights_all for computing optimal finite difference rules in a stable way.
Added higher order complex step derivative methods.
Version 0.7.7, December 18, 2014
 pbrod (35):
Got travisci working in order to run the tests automatically.
Fixed bugs in Dea class.
Fixed better error estimate for the Hessian.
Fixed tests for python 2.6.
Adding tests as subpackage.
Restructerd folders of numdifftools.
Version 0.7.3, December 17, 2014
 pbrod (5):
Small cosmetic fixes.
pep8 + some refactorings.
Simplified code by refactoring.
Version 0.6.0, February 8, 2014
 pbrod (20):
Update and rename README.md to README.rst.
Simplified call to Derivative: removed step_fix.
Deleted unused code.
Simplified and Refactored. Now possible to choose step_num=1.
Changed default step_nom from max(abs(x0), 0.2) to max(log2(abs(x0)), 0.2).
pep8ified code and made sure that all tests pass.
Version 0.5.0, January 10, 2014
 pbrod (9):
Updated the examples in Gradient class and in info.py.
Added test for vec2mat and docstrings + cosmetic fixes.
Refactored code into private methods.
Fixed issue #7: Derivative(fun)(numpy.ones((10,5)) * 2) failed.
Made print statements compatible with python 3.
Version 0.4.0, May 5, 2012
 pbrod (1)
Fixed a bug for inf and nan values.
Version 0.3.5, May 19, 2011
 pbrod (1)
Fixed a bug for inf and nan values.
Version 0.3.4, Feb 24, 2011
 pbrod (11)
Made automatic choice for the stepsize more robust.
Added easy to use interface to the algopy and scientificpython modules.
Version 0.3.1, May 20, 2009
 pbrod (4)
First version of numdifftools published on google.code
Copyright (c) 20092022, Per A. Brodtkorb, John D’Errico All rights reserved.
Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:
Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
Neither the name of the copyright holders nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS “AS IS” AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
Project details
Release history Release notifications  RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for numdifftools0.9.41py2.py3noneany.whl
Algorithm  Hash digest  

SHA256  a8b162e06889ea73643a47b84935a63d8214d4b4b0805d36a3c28c56379b3e51 

MD5  01116e9ed1a71bd724e8de0ab6c3bcc0 

BLAKE2b256  a35c37cd5db8c465db2664b2219410b8bc7743da6edb1b616b5d13008bd7cac2 