L4T-ML:r35.1.0-py3 broken Scikit-Learn libraries

[michael.janus]

When I import the sklearn libraries within the L4T-ML docker image (via jupyter notebook), I get an ImportError:
‘Contents of /usr/local/lib/python3.8/dist-packages/sklearn/__check_build/…/…/scikit_learn.libs…’ … It seems taht scikit-learn has not been built correctly.

As a potential fix, I tried to pip install sklearn in the docker, but that just reports that the package is already there and requirements are already there.

Furhtermore, what I find strange is that if I check the python version via ‘python --version’, it reports ‘2.7.18’ while the packages mention python 3.8 and jupyter kernel says ‘python 3’. Is 2.7.18 a special built for Jetpack that is actually a 3.x version???

[AastaLLL]

Hi,

We can import sklearn without issue.

Could you try it again or share the way to reproduce this?

Thanks.

[michael.janus]

hmm, that is weird, the sklearn import now succeeds, but the
import cv2
now gives ImportError /usr/lib/aarch64-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block

Maybe the sklearn issue pops up in specific combinations or order of imports, I’ll try a bit more to reproduce

[michael.janus]

Ok, after more trial and error I can reproduce the issue

from tensorflow.keras.utils import to_categorical
from sklearn.model_selection import train_test_split

[dusty_nv]

Hi @michael.janus, please see this thread:

OSError: /usr/lib/aarch64-linux-gnu/libgomp.so.1 ONLY IN JUPYTER NOTEBOOK

I am using the Jetbot 0.4.3 (with Jetpack 4.4.1). I am running the jupyterlab remote connection server. I get the error OSError: /usr/lib/aarch64-linux-gnu/libgomp.so.1: cannot allocate memory in static TLS block when trying to import torch inside jupyter notebook. I am aware that you can use: export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libgomp.so.1 to fix this issue but it doesn’t work for me. HOWEVER, I get no error if I import torch inside the Python CLI terminal. What Can I do?