Stefan Kroboth

> Looks like my state will have to carry on a NxP 2d array, where N is my population size and P is number of parameters. This way I'll be...

It is needed (at least) in [BFGS](https://github.com/argmin-rs/argmin/blob/61d9ada86a1d3be7bb545b0a88bce719b3b90808/argmin/src/solver/quasinewton/bfgs.rs#L149). Both [ndarray](https://docs.rs/ndarray/latest/ndarray/linalg/trait.Dot.html) and [np.dot](https://numpy.org/doc/stable/reference/generated/numpy.dot.html) compute the matrix product when given two 2D matrices. The `ArgminDot` implementation for Vecs should also do so (unless...

I've had a closer look and I agree, the transpose is complete nonsense. Not only that, the entire implementation makes no sense at all! This is quite embarrassing. I have...

I'd argue the current implementation is definitely wrong, even for BFGS ;) The only reason why nobody noticed/complained is likely because people tend to (and should) use the ndarray/nalgebra math...

This is huge! Thanks a lot, not just for CMA-ES, but also for the massive extension of argmin-math, which will certainly come in handy in the future. Since this is...

> The vector and nalgebra backend requires more iterations to converge because Eigenvector decomposition implementation is not as good a the one that comes with LAPACK Thanks for the explanation!...

> Hi, I pulled the changes from the branch `VolodymirOrlov/main` and modified the `examples/cmaes.rs` file to allow for 1000 iterations. The optimization still panics. Am I doing something wrong? Interestingly,...

From a quick look I've noticed that there are some direct comparisons of floats, for instance: ```rust if (e[i + 1] == 0.0) { eigenvalues[i + 1] -= u; e[m]...

Preconditioning would be a nice feature! If I understand it correctly, [it only requires another multiplication with a matrix (or the inverse of a matrix) with a vector](https://en.wikipedia.org/wiki/Conjugate_gradient_method#The_preconditioned_conjugate_gradient_method). I'm however...

Hi @Trombach, thanks a lot for this PR! :) At first glance it looks already quite good. Admittedly, I'm not familiar with this method. Could you provide a reference (paper,...