R. Padraic Springuel wrote:
I know its been a while since I last brought this up (Vol 19, Issue 1), but after some digging, I've discovered Scientific.Functions.Derivatives (it came with the Enthought Edition of python that I use). It defines a DerivVar type that works well for taking derivatives of functions, supporting partial derivatives (even cross partials) very well. I haven't fully tested the limits of its accuracy, but some basic tests show it to be at least as accurate as the top level derivative function in scipy. Perhaps instead of trying to modify derivative (and central_diff_weights) it'll be easier in future versions of scipy to include this type and raise it to the top level.
This is a nice class as it does automatic differentiation. I like a lot of what is in Scientific (though I don't like the naming conventions). But, does it handle all cases? What if the function is not built from standard functions? Does it perform numerical differentiation then? -Travis