Environment setup is broken due to pip/conda conflicts (scipy/sklearn/dask-ml)

[StRongTomatoes]

Hi,

First of all, thank you for the work you put into this project. I was really interested in trying it and exploring its potential.

However, after following the installation instructions from the README and recreating the environment from scratch, I ran into serious dependency issues that make the setup very difficult to use in practice.

The main problem seems to be that the environment mixes conda and pip for core scientific packages (numpy, scipy, scikit-learn, dask-ml). This leads to binary incompatibilities and broken imports.

For example, after installation:
python -c "import dask_ml, sklearn, scipy"

results in errors such as:
ImportError: cannot import name 'get_tags' from sklearn.utils
ImportError: cannot import name '_promote' from scipy.spatial.transform

This happens because pip-installed wheels conflict with conda-installed libraries, especially on Linux/HPC systems.
Unfortunately, after spending a significant amount of time trying to fix this, I decided to step back from using the project for now.

I believe this could be greatly improved by:

• Installing numpy, scipy, scikit-learn, and dask-ml only via conda-forge
• Leaving pip only for pure-Python packages
• Updating the .yml file accordingly

I hope this feedback is useful, because the project itself looks very promising.

Best regards,
StRongTomatoes

[Genereux-akotenou]

Hi @StRongTomatoes,

Thank you very much for taking the time to test MegaPlantTF and for providing such detailed feedback we truly appreciate it.

You are absolutely right that mixing pip and conda for core scientific packages can sometimes lead to binary incompatibilities. The current environment configuration mirrors the one used to deploy our webserver, which is why it was structured this way.

To help us reproduce and properly investigate the issue, could you please share more details about your system (Linux distribution, version, Python version, and Conda version)? This would allow us to better diagnose the conflict.

In parallel, we will review the .yml file and work on reducing the reliance on pip for compiled scientific packages to minimize the risk of such conflicts. We will push an updated and more robust environment configuration soon.

In the meantime, you can still use the online webserver to quickly test MegaPlantTF here:
https://bioinformatics.um6p.ma/MegaPlantTF

Thank you again for your feedback.

Best regards,
Généreux