I’m working on generalizing the embedded Laplace for cases where the likelihood admits a non-diagonal Hessian, but rather a block-diagonal Hessian (this ends up including the diagonal and fully dense cases).
Does Eigen support operations with block-diagonal matrices? In particular, I’m interested in cholesky decomposition, multiplication, and solve. A quick search didn’t return anything promising.
I have a good sense of how to write these things out, but I figured I’d check if the work had already been done.