from_scvi_model about scib-pipeline HOT 10 CLOSED

I think I revise the scib package, not the scib-pipeline.

And as for the GLIBCXX_3.4.30 error, I fixed it by doing this:
conda install libgcc
And if it doesn't work for you, you could try to find the location of lib, and I also put it in my bashrc:
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/data/msun/miniconda3/lib:/data/msun/miniconda3/envs/scib-pipeline-R4.0/lib

And then another error occured: libffi.so.7: cannot open shared object file: No such file or directory
It's because it finds the libffi.so.7.1.0, I copyed the correct one. You can also link one.
cp miniconda3/lib/libffi.so.7 miniconda3/envs/scib-pipeline-R4.0/lib/

Hope it can help you!

from scib-pipeline.

Hi!

We recently updated the scvi code in the repo (last week), so this must be a bug we missed. It sounds like just a parameter naming mismatch. Which version of scvi-tools do you have in your environment?

from scib-pipeline.

Hello, the version I have seems to be the latest. scvi-tools in /home/ubuntu/miniconda3/envs/scib-pipeline-R4.0/lib/python3.7/site-packages (0.16.4). I installed it using conda: install scvi-tools -c conda-forge as the first error I encountered when testing the snakemake was that scvi was not installed. However, I wonder if the python version is the issue. When I start python and try to import the scvi, it gives me an error to do GLIBCXX_3.4.30. I have tried many ways to update to GLIBCXX_3.4.30, but haven't been able to. But then, equally, not sure why it runs when I start the snakemake pipeline and only dies at the end. The exact import error if I try to import it as standalone is: ImportError: /lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.30' not found (required by /home/ubuntu/miniconda3/envs/scib-pipeline-R4.0/lib/python3.7/site-packages/jaxlib/xla_extension.so)
Thank you so much for taking the time.

from scib-pipeline.

Also maybe related or unrelated. If I set up a new environment for scvi as in the docs: conda create -n scvi-env python=3.9, but I install with conda forge, I get import problems. I can avoid those problems by using pip install scvi-tools (someone posted somewhat similar issues with h5py). So I will try to install scvi with pip in the scib-pipeline-R4.0 just to check. The error when I use conda forge: (ImportError: /home/ubuntu/miniconda3/envs/scvi-env/lib/python3.9/site-packages/h5py/defs.cpython-39-x86_64-linux-gnu.so: undefined symbol: H5Pset_fapl_ros3).

from scib-pipeline.

Hi, I also got the same error. I tried to revise this file: "unknown_category" to "unlabeled_category".

 miniconda3/envs/scib-pipeline-R4.0/lib/python3.7/site-packages/scib/integration.py", line 281, in scanvi
     unknown_category="UnknownUnknown",  # pick anything definitely not in a dataset

And it seems to work. It goes like this:

 snakemake --configfile configs/test_data-R4.0.yaml --cores 24
Building DAG of jobs...
 Nothing to be done (all requested files are present and up to date).
 Complete log: .snakemake/log/2022-06-29T155249.803624.snakemake.log

Do you fix it by installing scvi-tools by pip?

from scib-pipeline.

This is a great discussion thread! @littleju714 would you be up for creating a small PR with the fix?

@mumichae we might need to change the scvi install in the env files if the pip install works better. Maybe the version in conda-forge is not maintained? Would be good to check with @agayoso.

from scib-pipeline.

Hello, I tried with pip, but it takes me in a rabbit hall of conflicts with other packages and with tensorflow so I couldn't get it working in the end. Maybe if I change the "unknown_category" to "unlabeled_category"? Will try that. Thank you again.

from scib-pipeline.

Thank you, will try that and also isolating in a separate env as it has been suggested.

from scib-pipeline.

Related: theislab/scib#324

from scib-pipeline.

There have been some updates in dependencies recently. I will close this thread for now. Please open up a new issues for any problems that still persist

from scib-pipeline.