cannot install using environment.yml

broadinstitute / grit-benchmark

Benchmarking a metric used to evaluate a perturbation strength

BSD 3-Clause "New" or "Revised" License

5 stars 5 forks source link

cannot install using environment.yml #34

Open alxndrkalinin opened 1 year ago

alxndrkalinin commented 1 year ago

Running conda env create -f environment.yml returns PackageNotFound for dvc and dvc-s3. If those are moved to the pip section, it complains about not finding s_gd2 that is used by phate.

Seems like it's Apple Silicon-related issues and environment.yml should be updated to address them and fix installation.

alxndrkalinin commented 1 year ago

Turns out some packages do not have conda binaries compiled for Apple Silicon. This can be fixed by installing them via pip: 8c4224287c4d9b1ec4a73a468f6e2118b40d081b

However, there are issues with installing packages for perturbseq preprocessing from [perturbseq_processing_environment.yml]() Specifically, bioconductor-biostrings and r-seurat also do not have apple silicon binaries. Installing everything else and trying to install these two directly from inside R fails as there are issues with compiling the source code.

alxndrkalinin commented 1 year ago

I managed to install from perturbseq_processing_environment.yml by:

removing r-seurat and bioconductor-biostrings from yml, bumping nb_conda_kernels to 2.3.1 and installing from yml with conda
removing MKL from conda environment as described in https://github.com/satijalab/seurat/issues/3273#issuecomment-660654642
setting compiling flags as described in https://github.com/conda-forge/r-base-feedstock/issues/163#issuecomment-989515053

finally, installing in R console:


install.packages('sctransform')
install.packages('Seurat')

if (!require("BiocManager", quietly = TRUE)) install.packages("BiocManager")

BiocManager::install("Biostrings")

alxndrkalinin commented 1 year ago

Bumped dependency versions for creating figure environment. Commented it out r-magick, because it does not currently have osx-arm64 binaries and was only used to read a pre-made figure panel.

mfansler commented 1 year ago

We now have r-seurat building for osx-arm64 on Conda Forge. But bioconductor-biostrings is a different story, since those are built by Bioconda, which has not yet set up CI for osx-arm64 builds.

alxndrkalinin commented 1 year ago

Thanks for letting us know @mfansler! I just tested it and it works perfectly. Turns out, we were not using bioconductor-biostrings, so everything installs fine through conda now.