Kernel crashed when executing gpboost code

[melantha-wang]

Hi Fabio,

Thanks for the awesome package! I just started to play around with the Python package and when running this code example in jupyter I kept receiving this error message:

The Kernel crashed while executing code in the the current cell or a previous cell. Please review the code in the cell(s) to identify a possible cause of the failure. Click [here](https://aka.ms/vscodeJupyterKernelCrash) for more info. View Jupyter [log](command:jupyter.viewOutput) for further details.
Canceled future for execute_request message before replies were done

I was able to import the library but Python would crash as soon as I run this line:

gp_model = gpb.GPModel(group_data=group, likelihood="gaussian")

Do you have any idea what could have caused this? I'm working on a M1 machine in case that can be relevant.

[fabsig]

Thank you for reporting this!

There is currently no pre-compiled version on PyPI for the Mac M1. This means that you have to to do the compilation yourself. See the discussion #51 for more details.

[fabsig]

FYI: The R package has a pre-compiled macOS arm64 on CRAN.

[melantha-wang]

Thanks for your reply!

Unfortunately I'm still experiencing the same issue after building the package (using Apple Clang and cmake) and installing from the source. I didn't get any error message during the installation process so it all seemed to work fine but the code just didn't run. The R package is a workable solution but I would love to know if there's anything else I can do to resolve this. Thanks!

[fabsig]

Thank you for your update.

Unfortunately, I cannot reproduce this. I have tried installing the gpboost Python package on a M1 Mac mini on scaleway and it works without any problems for me both using clang or gcc after following these instructions.

@estsauver once mentioned in #51 the option to fund a scaleway instance for GPBoost. Unfortunately, I have not heard anything so far. Once this is set up, I can provide a pre-compiled macOS arm64 version on PyPI.

[fabsig]

@melantha-wang: Just to double check that it is not an issue with the sklearn interface: does this code work for you?

[melantha-wang]

Unfortunately that didn't work either. It crashes as soon as I run gp_model = gpb.GPModel(likelihood="gaussian", group_data=np.array([0, 1, 2])).

One thing I just noticed was that if I installed from source, I can only see the egg-info (gpboost-0.7.7-py3.9.egg-info) and not the actual package in the site-packages/ directory. I did successfully pip installed the package before so didn't notice this until I uninstalled and reinstalled. Can that be the issue?

[fabsig]

Yes, this is probably the reason. You need to have a gpboost directory with a lib_gpboost.so file in it. This is the file where all the compiled C++ functions are. Your code crashes when Python tries to call this the first time (and probably does not find it). You might want to uninstall and re-install. Also make sure that the installation including the compilation ends successfully.

[fabsig]

A macOS arm64 chip pre-compiled version is now on PyPI. You should now be able to install with pip on M1 / M2 Macs without any problems as compilation is not required anymore.

[melantha-wang]

Thanks for the update! I'm still having the same issue of kernel crashing though... (I can install & import successfully but the kernel crashed every time I tried to call a function within the package).

[fabsig]

Thank you for your feedback. Two things that might help:

• In case you migrated from an older Mac, you need to uninstall homebrew and reinstall it before you install libomp (to make sure that the arm64 version is used)
• Use Miniforge (brew install miniforge and conda init "$(basename "${SHELL}")") instead of say Anaconda

With this, it it runs on my arm64 Mac. I hope this helps.