Peter Andreas Entschev

The idea to use `normalize_to_array()` sounds good to me. I'm now wondering about the case where we need to convert these outputs back to their original type arrays. We can...

It's not ok in here: https://github.com/dask/dask-glm/blob/master/dask_glm/algorithms.py#L347 We pass `beta0`, `X` and `y` to `scipy.optimize.fmin_l_bfgs_b`. `beta0` gets has to be copied back to CPU because it gets modified here: https://github.com/scipy/scipy/blob/master/scipy/optimize/lbfgsb.py#L326 Later...

I actually had the exact same idea with the `y[:] = x` assignment before, but I assumed the non-automatic conversion was intentional. I think if we can allow that to...

I found two open issues related to what we just talked here: https://github.com/cupy/cupy/issues/593 https://github.com/cupy/cupy/issues/589

> Minor comment: could you please add an `n_iter_` attribute to the estimator docstrings Yes, thanks for pointing that I forgot it. > Also, as you mentioned, these changes will...

From my side, this is ready for another review/merge.

Are we good to merge this?

> I think that one of the reasons that the first line turns me off is that it instructs people to do something. I would be more comfortable if it...

Besides the above I was looking at the code in more detail and I see the following block: https://github.com/NVIDIA-Merlin/core/blob/6e52b48140615708b59926b5f9c3601f8feeab93/merlin/core/compat/__init__.py#L102-L105 This is creating a new context on a GPU only to...

Although this is tripping that block, I would suggest always using PyNVML to query GPU information, specifically what I mention in https://github.com/NVIDIA-Merlin/core/issues/363#issuecomment-1888595036 can be dangerous with Dask if for some...