Registered Member
|
Hi.
I have a project that could benefit from automatic differentiation. I see that there are two Eigen modules for AD: Adolc and Autodiff. I know that both modules are unsupported, but I am hoping to get a recommendation for which module to use. The functions that I need to evaluate can be quite complicated, and often call other self-cobbled functions that take Eigen matrices as arguments. So I would like to avoid rewriting as much of my existing Eigen code as I can. Is either approach thought to be easier to implement than the other? How does the runtime performance compare? Also, how would I go about supplying derivatives of functions that are not intrinsically built in. For example, I often need to differentiate the log gamma function. What would I need to do to tell Adolc or Autodiff how to differentiate that? Will the modules accept Eigen plugins? And finally, would someone be able or willing to share an example of either Adolc or Autodiff in action? I'm looking for something perhaps a bit more complex than the test examples (e.g., differentiating functions that call other functions). Thanks, as always, for your help. Michael |
Registered users: Bing [Bot], Google [Bot], Yahoo [Bot]