Wessel

> While it's not too hard to contrive a Hermitian matrix with exactly repeating eigenvalues (e.g. a diagonal matrix), I've had no luck so far coming up with a non-Hermitian...

@willtebbutt That sounds very sensible to me. How would `f1.mul_approx(f2)` or `f1.mul_matched(f2)` sound? Do you perhaps have a suggestion for a name?

Perfect! Let's go with `mul_matched` then. I'll incorporate the change in the 1.0 release.

Hi @vdsmax! I've put together a simple MOGP model (not using the example) which might better suit your use case. The script uses JAX to learn hyperparameters. (You can also...

Ouch! That doesn't look good. Could you confirm that running other JAX code on the GPU works fine? If that's the case, I can look into this more closely to...

Hey @vdsmax, That's very frustrating. I'm not sure what's going wrong. I am able to run the example on my end on a GPU. I am running `jaxlib-0.1.73+cuda11.cudnn82` and `jax-0.2.25`....

If I'm not mistaken, the parameter `eps` is used to calculate the round-off error and should be set to the machine epsilon of the corresponding data type. The issue is...

Yes, that's a good suggestion. We should carefully check the code, to ensure that the assumption of `Float64`s is not also subtly baked in in other places.

Yes, that's an even better solution.

Do we have a good example where the accuracy for `Float32`s is not what we would like it to be? I wonder if v0.12 resolves these issues.