opengeophysics / matrixutils

Utilities for working with matrices as linear operators
MIT License
1 stars 0 forks source link

Finite Difference Sensitivity #3

Open lheagy opened 6 years ago

lheagy commented 6 years ago

Over in opengeophysics/scratchpad, @prisae and I have been working on using SimPEG to invert empymod simulations. One piece we wrote to facilitate this is a util to approximate the sensitivity of a function using a finite difference approximation. This might fit in matrixutils? I would appreciate thoughts on scope here. cc @leouieda, @prisae, @rowanc1

prisae commented 6 years ago

Not sure. It wouldn't be my first thought to search a gradient function in a package called matrixutils. But then, the Jacobian usually is a matrix. I guess it would fit.

prisae commented 6 years ago

Or would it fit better into https://github.com/opengeophysics/deeplook? @leouieda

rowanc1 commented 6 years ago

Any chance we can just rely on this: https://docs.scipy.org/doc/scipy-0.14.0/reference/generated/scipy.optimize.approx_fprime.html

I sort of agree with @prisae that this is weird to be in the matrixutils repo, seems like it fits with inversion or optimization much more.

leouieda commented 6 years ago

I agree that this could be probably fit in better with the inversion packages. I think PyGIMLi has something like this as well.