Installing ASReview backend fails on Apple M1 chip

[ottomattas]

Describe the bug
Trying to install ASReview in DEVELOPMENT mode fails on an Apple M1 chip. Process takes about 1 hour to complete.

To Reproduce
Steps to reproduce the problem:

1. Pull the latest code from the repository
2. Navigate to the repository locally
3. pip install -e .

Version information

• OS: macOS Big Sur Version 11.5.2 (20G95)
• MacBook Pro (13-inch, M1, 2020)

Additional context
Full CLI output attached as a zip container.

[ottomattas]

From the looks of it, there are three packages that need a workaround - h5py, scikit-learn, scipy.

Relevant section from the end of CLI output:

Successfully built gevent greenlet pandas zope.interface
Failed to build h5py scikit-learn scipy
Installing collected packages: scipy, numpy, Jinja2, itsdangerous, click, zope.interface, zope.event, scikit-learn, pytz, python-dateutil, greenlet, flask, et-xmlfile, xlrd, sklearn, rispy, pandas, openpyxl, h5py, gevent, flask-cors, dill, asreview

[ottomattas]

For users

There is a quick workaround, using miniforge.

# Install miniforge
% brew install miniforge

# Create environment
% conda create --name asreview_m1

# Initialise (zsh) shell
% conda init zsh

# Source (zsh) shell directly
% . ~/.zshrc

# Activate environment
% conda activate asreview_m1

# Install packages which fail to install via pip
% conda install h5py scipy scikit-learn

# Install asreview
% pip install asreview

# Delete the environment, in case you need to clean up afterwards
% conda deactivate
% conda env remove -n asreview_m1

This workaround was inspired by Hendrik Erz, a researcher and an Open Source contributor. If you want to read up on the method proposed here, you can see his blog post on the issue.

For developers

There is a quick workaround, using miniforge.

# Install miniforge
% brew install miniforge

# Create environment
% conda create --name asreview_m1

# Initialise (zsh) shell
% conda init zsh

# Source (zsh) shell directly
% . ~/.zshrc

# Activate environment
% conda activate asreview_m1

# Install packages which fail to install via pip
% conda install h5py scipy scikit-learn

# Install asreview
% pip install -e .

# Delete the environment, in case you need to clean up afterwards
% conda deactivate
% conda env remove -n asreview_m1

This workaround was inspired by Hendrik Erz, a researcher and an Open Source contributor. If you want to read up on the method proposed here, you can see his blog post on the issue.

[MarielleZZ]

Starting with a git clone + cd on asreview and proceeding with the workaround above was successful for me:

git clone https://github.com/asreview/asreview.git
cd asreview

brew install miniforge
conda create --name asreview_m1

conda init zsh
. ~/.zshrc
conda activate asreview_m1
conda install h5py scipy scikit-learn
pip install -e .
conda deactivate
conda env remove -n asreview_m1

[ottomattas]

Thanks for reporting back and confirming the proposed solution @MarielleZZ!

[hjiajing]

I didn't address this M1 issue following the above workaround. But I ran pip3 install asreview and it worked fine. Then I ran asreview lab, the asreview ran successfully. 😂

(base) aaronhu@AarondeMacBook-Air ~ % python --version
Python 3.9.7
(base) aaronhu@AarondeMacBook-Air ~ % pip3 --version
pip 21.2.4 from /Users/aaronhu/opt/miniconda3/lib/python3.9/site-packages/pip (python 3.9)

[pietervandekerckhove]

Thank you @hjiajing , it worked for me as well in this way :)