Unconstrained optimization with trust newton

[Praharsh Suryadevara]

Hi, 

We have a non-convex function to optimize in 100-2000ish dimensions, where we want solutions that are close to the original point, and where time is a limiting factor (but we have analytical hessians).

I was thinking of using the algorithm mentioned in [ByrdSchnabel] (since Nocedal Wright said that it helps with the indefinite hessian case), which seems to be implemented in Ceres as the SUBSPACE_DOGLEG method (?).

 The full hessian isn't mentioned in the general unconstrained optimization section of the documentation, so I wanted to check whether there is any way of using the trust region methods on a full Newton step that is not mentioned in the documentation through the package.

[ByrdSchnabel] Byrd, Richard H., Robert B. Schnabel, and Gerald A. Shultz. "Approximate solution of the trust region problem by minimization over two-dimensional subspaces." Mathematical programming 40.1 (1988): 247-263.

[Sameer Agarwal]

I am sorry but that is not possible.

--
You received this message because you are subscribed to the Google Groups "Ceres Solver" group.
To unsubscribe from this group and stop receiving emails from it, send an email to ceres-solver...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/ceres-solver/6082b97d-38df-438a-96b9-ecdc4b7550d7n%40googlegroups.com.