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()
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
[JPerktold14]  Perktold, J (2014), numdiff package http://statsmodels.sourceforge.net/0.6.0/_modules/statsmodels/tools/numdiff.html 
[Ver14]  Adriaen Verheyleweghen, (2014) “Computation of higherorder derivatives using the multicomplex step method”, Project report, NTNU 
[GLD12]  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 
[MELEV12]  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. 
[Lan10]  Gregory Lantoine (2010), “A methodology for robust optimization of lowthrust trajectories in multibody environments”, Phd thesis, Georgia Institute of Technology 
[Rid09]  Ridout, M.S. (2009) “Statistical applications of the complexstep method of numerical differentiation”, The American Statistician, 63, 6674 
[DErrico06]  D’Errico, J. R. (2006), “Adaptive Robust Numerical Differentiation”, http://www.mathworks.com/matlabcentral/fileexchange/13490adaptiverobustnumericaldifferentiation 
[KLLK05]  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. 
[For98]  B. Fornberg (1998) “Calculation of weights_and_points in finite difference formulas”, SIAM Review 40, pp. 685691. 
[For81]  Fornberg, B. (1981). “Numerical Differentiation of Analytic Functions”, ACM Transactions on Mathematical Software (TOMS), 7(4), 512526. http://doi.org/10.1145/355972.355979 
[JML69]  Lyness, J. M., Moler, C. B. (1969). “Generalized Romberg Methods for Integrals of Derivatives”, Numerische Mathematik. 
[JML66]  Lyness, J. M., Moler, C. B. (1966). “Vandermonde Systems and Numerical Differentiation”, Numerische Mathematik. 
[NAG]  NAG Library. NAG Fortran Library Document: D04AAF 
Changelog
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) 20092021, 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 numdifftools0.9.40py2.py3noneany.whl
Algorithm  Hash digest  

SHA256  57bb19ba0abc914ad8373ad48a1416f186bf0a24dc677f138350b26895301d40 

MD5  6c35181120a95eadb9892ec0aa559c4e 

BLAKE2256  7e44b985d36bf97ca3af694be9dee299716b97cc43a6184be7a6e831350c6cb6 