Kristof Schröder

@sleepymalc Thank you for reporting. Could you please provide more information, so we can reproduce and fix the issue? The easiest thing would be a minimal, reproducible example. Which implementation...

@sleepymalc thanks, that helped a lot. Please have a look at #570. To install via pip ``` pip install git+https://github.com/aai-institute/pyDVL.git@fix/569-missing-device-move-ekfac ``` To test it, I added a small call to...

@sleepymalc awesome, please let us know, if you encounter any other issues, thanks:)

@jakobkruse1 I would not suggest, that we modify the code for every architecture. I would rather try except on the RuntimeError and make a meaningful error message, maybe even referencing...

Potentially resolved by #558

@janosg potentially resolved by #558

The reason for the inconsistent test behavior is that: ```python np.array([np.nan]).astype(int)[0] == -9223372036854775808 (on linux-amd64 and maybe other systems) np.array([np.nan]).astype(int)[0] == 0 (on osx-arm64) ``` see also [this numpy issue](https://github.com/numpy/numpy/issues/21166)....

The reason for the occurrence of `np.nan`is the following code: ```python class ClosedFormLinearClassifier: def __init__(self): self._beta = None def fit(self, x: NDArray, y: NDArray) -> float: v = x[:, 0]...

@kosmitive do you have time to support on this? I created a draft PR with a temporary fix.

Usage of MapReduce is deprecated, fine grain config is achieved in #558 via joblib.