On Sun, Oct 30, 2011 at 11:03 AM, <josef.pktd@gmail.com> wrote:
2011/10/30 François Boulogne <boulogne.f@gmail.com>:
-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1
Dear all,
I was wondering if a piece of code has been developped for derivative calculations, espacially for a sample (array of points) like for integration: http://docs.scipy.org/doc/scipy/reference/tutorial/integrate.html with different methods (right or left first derivatives, second
derivatives...)
I didn't succeed in finding this in the documentation. Does it exist? If not, is it planned by someone?
There is one helper function in scipy.optimize (scipy.optimize.optimize), nothing else in scipy.
Well, not exactly "nothing else"... (eat's email arrived as I was typing this, so it will echo some of what he said.) Functions that operate on a discrete sample: numpy.diff This can be used to compute a derivative by dividing by the appropriate power of dx. numpy.ediff1d Like numpy.diff, but strictly for 1D arrays. It also provides the option for specifying values to append to the ends of the array before computing the difference. numpy.gradient Return the gradient of an n-d array. scipy.fftpack.diff Derivative of a periodic sequence. See http://www.scipy.org/Cookbook/KdV for an example. Functions that operate on a callable function: scipy.misc.derivative Find the n-th derivative of a function at point x0. scipy.misc.central_difference_weights Return weights for an Np-point central derivative scipy.optimize.approx_fprime No docstring (sigh), but from the source code (use approx_fprime?? in ipython), it is pretty easy to figure out what it does. Having said that, I think a module specifically for computing derivatives (with good docs and tests), as being discussed in the ticket #1510 ( http://projects.scipy.org/scipy/ticket/1510) would be a nice addition. Warren
My standard recommendation for finite differences is numdifftools, it's on pypi. There is a ticket that asks for it's inclusion in scipy, IIRC.
There are some packages on automatic differentiation.
(and we have our own hacked together numdiff in statsmodels, just for optimization and Hessian calculations.)
Josef
Thanks. Cheers,
- -- François Boulogne.
Membre de l'April - Promouvoir et défendre le logiciel libre http://www.april.org -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.11 (GNU/Linux)
iQIcBAEBAgAGBQJOrXHDAAoJEKkn15fZnrRLxbYP/3/NUzU5metNhGc4tw6Uo0mw kIlAxf1cvtopbJ+JTXrmKkJpyYUI1bMsqVKpPniW073dgPAfAX3Bajymysh5Hgnd 5+0hh7v8JmsIvKm9HrEePSrENYrIVTfFyvRV+tBLhkfHJ9Vj7uUy7a3/lyRS6s7v FEItXLhkNYtqEir8h2eZ7uW178mwq6nBl6Zi5UOjOXq7u0SxcnusKkYyiL+CirSN KafcrU+rUdHw6khE1exj435GMSKx+N3+rV/kDAQWWQc0ncWmdX2Jr7PapT1DQbNN b3UK2NjYGDZE2NMRuZmRmeTIk2S+PVPqRxEu0x62CS4Y4JnTaZx1xSiPGP0d9cZC NG4/9h0LFM8rxj/kxYjBakbIs6LhwPnS1ZvYf00o3eaY4+tLGx3UozoLpEWSCOuQ GrONTWvdUlfVWsd3qOhwCj+NORkz9yXJNqJHulgz6T8fKnMyPL+R80XfRuS1alxj m166ySHxrPDzwiSlV5vWR/1ajrEMAI/GYev55dtzlIx6hxfLVCOryFei/LNtlrN+ lYK48QqateqvLDs+NUq+7azmCAkc8fW5UeD7bBLdoGwlXvYaOEZh8+lzxU2sc0T4 93kNZAzOzv/CdcI1uhnp2NP8ODGEyRLx5cUlcKqmF/iAqtM2zc5Zca9YiygR2w83 XQsGdZYw4rkVMnUTk6Ep =/Nlz -----END PGP SIGNATURE-----
_______________________________________________ SciPy-User mailing list SciPy-User@scipy.org http://mail.scipy.org/mailman/listinfo/scipy-user
_______________________________________________ SciPy-User mailing list SciPy-User@scipy.org http://mail.scipy.org/mailman/listinfo/scipy-user