[Numpy-discussion] Calculation of a hessian

Eelco Hoogendoorn hoogendoorn.eelco at gmail.com
Fri Aug 8 10:37:48 EDT 2014


Do it in pure numpy? How about copying the source of numdifftools?

What exactly is the obstacle to using numdifftools? There seem to be no
licensing issues. In my experience, its a crafty piece of work; and
calculating a hessian correctly, accounting for all kinds of nasty floating
point issues, is no walk in the park. Even if an analytical derivative
isn't too big a pain in the ass to implement, there is a good chance that
what numdifftools does is more numerically stable (though in all likelihood
much slower).

The only good reason for a specialized solution I can think of is speed;
but be aware what you are trading it in for. If speed is your major concern
though, you really cant go wrong with Theano.

http://deeplearning.net/software/theano/library/gradient.html#theano.gradient.hessian
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/numpy-discussion/attachments/20140808/b9d5219c/attachment.html>


More information about the NumPy-Discussion mailing list