[Bug] Segfault on Spectral mixture GP regression test (MacOS)

🐛 Bug

Running the test/examples/test_spectral_mixture_gp_regression.py reliably produces a segfault on my Mac, even when installing into a virgin conda env. This happens on master but also on commits a few months old (I haven’t run tests locally in a while though, so not sure when this broke).

I don’t see this issue on Linux.

To reproduce

** Code snippet to reproduce **

conda create -n gptt python=3.8
conda activate gptt
conda install -c pytorch pytorch
conda install scipy scikit-learn

cd gpytorch
pip install -e .

(gptt) balandat@balandat-mbp gpytorch % pytest test/examples/test_spectral_mixture_gp_regression.py
==================================================================================================================== test session starts =====================================================================================================================
platform darwin -- Python 3.8.12, pytest-6.2.4, py-1.10.0, pluggy-0.13.1
rootdir: /Users/balandat/Code/gpytorch
collected 2 items

test/examples/test_spectral_mixture_gp_regression.py Fatal Python error: Segmentation fault

Thread 0x000000011e4aee00 (most recent call first):
  File "/Usezsh: segmentation fault  pytest test/examples/test_spectral_mixture_gp_regression.py

Here is the conda env:

# packages in environment at /Users/balandat/miniconda3/envs/gptt:
#
# Name                    Version                   Build  Channel
attrs                     21.2.0             pyhd3eb1b0_0
blas                      1.0                         mkl
ca-certificates           2021.10.26           hecd8cb5_2
certifi                   2021.10.8        py38hecd8cb5_0
gpytorch                  1.5.1                     dev_0    <develop>
iniconfig                 1.1.1              pyhd3eb1b0_0
intel-openmp              2021.4.0          hecd8cb5_3538
joblib                    1.1.0              pyhd3eb1b0_0
libcxx                    12.0.0               h2f01273_0
libffi                    3.3                  hb1e8313_2
libgfortran               3.0.1                h93005f0_2
libuv                     1.40.0               haf1e3a3_0
llvm-openmp               12.0.0               h0dcd299_1
mkl                       2021.4.0           hecd8cb5_637
mkl-service               2.4.0            py38h9ed2024_0
mkl_fft                   1.3.1            py38h4ab4a9b_0
mkl_random                1.2.2            py38hb2f4e1b_0
more-itertools            8.10.0             pyhd3eb1b0_0
ncurses                   6.3                  hca72f7f_1
numpy                     1.21.2           py38h4b4dc7a_0
numpy-base                1.21.2           py38he0bd621_0
openssl                   1.1.1l               h9ed2024_0
packaging                 21.0               pyhd3eb1b0_0
pip                       21.2.4           py38hecd8cb5_0
pluggy                    0.13.1           py38hecd8cb5_0
py                        1.10.0             pyhd3eb1b0_0
pyparsing                 3.0.4              pyhd3eb1b0_0
pytest                    6.2.4            py38hecd8cb5_2
python                    3.8.12               h88f2d9e_0
pytorch                   1.10.0                  py3.8_0    pytorch
readline                  8.1                  h9ed2024_0
scikit-learn              1.0.1            py38hae1ba45_0
scipy                     1.7.1            py38h88652d9_2
setuptools                58.0.4           py38hecd8cb5_0
six                       1.16.0             pyhd3eb1b0_0
sqlite                    3.36.0               hce871da_0
threadpoolctl             2.2.0              pyh0d69192_0
tk                        8.6.11               h7bc2e8c_0
toml                      0.10.2             pyhd3eb1b0_0
typing_extensions         3.10.0.2           pyh06a4308_0
wheel                     0.37.0             pyhd3eb1b0_1
xz                        5.2.5                h1de35cc_0
zlib                      1.2.11               h1de35cc_3

System information

Please complete the following information:

• master
• 1.10
• MacOS 11.6.1

[pipme]

FYI, several days ago I also encountered a segmentation fault on Mac M1 (conda rosetta) when reaching here https://github.com/cornellius-gp/gpytorch/blob/ade5db8df1f8e4acbc64ec1dd8809e58506827f1/gpytorch/kernels/spectral_mixture_kernel.py#L205-L207 but it was fixed after pip reinstalling scikit-learn. My current version is scikit-learn==1.0.1.

[wjmaddox]

Yeah, let’s just remember to visit this 3-6 months down the road and check scikit-learn again since the fix seems to be in their nightly.