Autokeras trying to import libgomp-d22c30c5.so.1.0.0 from scikit-learn on arm64/aarch64

Bug Description

I am trying to use autokeras on a Raspberry Pi 4B with 64-bit Raspberry Pi OS. It can be installed but would generate the following error on import.

Seems like it was trying to import libgomp-d22c30c5.so.1.0.0 from scikit-learn but no such file exists anywhere in the system. Both Tensorflow and scikit-learn can be imported and works properly on their own.

…However, this bug can be bypassed by import sklearn before import autokeras in the same script. Some error about Placeholder_5 would pop up but otherwise the model finished like it should.

Bug Reproduction

>>> import autokeras
Traceback (most recent call last):
  File "/usr/local/lib/python3.7/dist-packages/sklearn/__check_build/__init__.py", line 44, in <module>
    from ._check_build import check_build  # noqa
ImportError: /usr/local/lib/python3.7/dist-packages/sklearn/__check_build/../../scikit_learn.libs/libgomp-d22c30c5.so.1.0.0: cannot allocate memory in static TLS block

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "<pyshell>", line 1, in <module>
  File "/usr/local/lib/python3.7/dist-packages/autokeras/__init__.py", line 15, in <module>
    from autokeras.auto_model import AutoModel
  File "/usr/local/lib/python3.7/dist-packages/autokeras/auto_model.py", line 25, in <module>
    from autokeras import blocks
  File "/usr/local/lib/python3.7/dist-packages/autokeras/blocks/__init__.py", line 17, in <module>
    from autokeras.blocks.basic import BertBlock
  File "/usr/local/lib/python3.7/dist-packages/autokeras/blocks/basic.py", line 19, in <module>
    from kerastuner.engine import hyperparameters
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/__init__.py", line 20, in <module>
    from . import oracles
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/oracles/__init__.py", line 16, in <module>
    from ..tuners.bayesian import BayesianOptimizationOracle as BayesianOptimization
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/tuners/__init__.py", line 17, in <module>
    from .bayesian import BayesianOptimization
  File "/usr/local/lib/python3.7/dist-packages/kerastuner/tuners/bayesian.py", line 5, in <module>
    from sklearn import exceptions
  File "/usr/local/lib/python3.7/dist-packages/sklearn/__init__.py", line 81, in <module>
    from . import __check_build  # noqa: F401
  File "/usr/local/lib/python3.7/dist-packages/sklearn/__check_build/__init__.py", line 46, in <module>
    raise_build_error(e)
  File "/usr/local/lib/python3.7/dist-packages/sklearn/__check_build/__init__.py", line 41, in raise_build_error
    %s""" % (e, local_dir, ''.join(dir_content).strip(), msg))
ImportError: /usr/local/lib/python3.7/dist-packages/sklearn/__check_build/../../scikit_learn.libs/libgomp-d22c30c5.so.1.0.0: cannot allocate memory in static TLS block
___________________________________________________________________________
Contents of /usr/local/lib/python3.7/dist-packages/sklearn/__check_build:
_check_build.cpython-37m-aarch64-linux-gnu.so__pycache__               __init__.py
setup.py
___________________________________________________________________________
It seems that scikit-learn has not been built correctly.

If you have installed scikit-learn from source, please do not forget
to build the package before using it: run `python setup.py install` or
`make` in the source directory.

If you have used an installer, please check that it is suited for your
Python version, your operating system and your platform.

Data used by the code:

Expected Behavior

It would be nice if I don’t have to do the additional import before autokeras.

Setup Details

This is how I installed all the stuff on a fresh RPi OS: (I can’t say if all of it is necessary, but right now it works)

sudo apt-get install build-essential python3-dev python3-pip python-h5py python3-h5py libhdf5-dev libblas-dev liblapack-dev libopenblas-dev libatlas-base-dev gfortran -y

sudo pip3 install cython pip setuptools wheel --upgrade

sudo pip3 install numpy==1.18.5

sudo pip3 install wrapt --upgrade --ignore-installed

sudo pip3 install https://github.com/sanjayseshan/tensorflow-aarch64/releases/download/v2.4.0/tensorflow-2.4.0-cp37-cp37m-linux_aarch64.whl

sudo pip3 install autokeras

The aarch64 wheel for autokeras would install/build scipy, pandas and scikit-learn on its own.

• OS type and version: Debian GNU/Linux 10 (buster), kernel: Linux 5.4.79-v8+, architecture: arm64
• Python: 3.7.3 (64 bit)
• autokeras: 1.0.12
• keras-tuner: 1.0.2
• scikit-learn: 0.24.0
• numpy: 1.18.5
• scipy: 1.5.4
• pandas: 1.1.5
• tensorflow: 2.4.0

Hardware is Raspberry Pi 4B 4GB. The 64-bit OS was downloaded from https://ftp.ubuntu-tw.net/mirror/raspbian-downloads/raspios_arm64/images/ and updated/upgraded before installing all the packages above.

Additional context

[Sonins]

I resolved this problem by setting environment variable LD_PRELOAD to path to libgomp-d22c30c5.so.1.0.0.

export LD_PRELOAD='/PATH/TO/libgomp-d22c30c5.so.1.0.0'

For this particular issue,

export LD_PRELOAD='/usr/local/lib/python3.6/dist-packages/scikit_learn.libs/libgomp-d22c30c5.so.1.0.0'

And execute program again.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.