Are there any Julia packages which do high-dimensional unconstrained optimization and support diagonal-only updates to the Hessian? Those that I know (Optim and OptimKit) have BFGS which updates the entire Hessian, but here I’m looking to just update the diagonal (since, for various reasons, my off-diagonals end up very numerically noisy and I pretty much know they should be small anyway).
It might not be hard to hack either of those packages and take only the diagonal of the BFGS update, but apparently that’s actually not the optimal choice (e.g. as stated here, sorry for paywall, but you can see this statement in the abstract). Thanks for any tips!