Open erzakiev opened 1 month ago
it seems like (for now, at least), when taking the first route but using python v3.9, i.e.
conda create -n scArches # creating new environment
conda install python=3.9 # installing the recommended python version
conda install anaconda::jupyter
pip install -U scarches
it doesn't complain when importing scPoli
anymore. But pip was complaining about incompatibility of phate and pandas was uninstalled, which is concerning:
So I was able to run the code in what seems to be - judging by the speed of training - a cpu mode. But when trying train, for instance, an scGen
model - it takes ages. I suspect that it's because on macos there's no pytorch-cuda
, but we do have mps
.
When running, for instance, a simple training from the tutorial,
network = sca.models.scgen(adata = source_adata,
hidden_layer_sizes=[256,128])
network.train(n_epochs=epoch, early_stopping_kwargs = early_stopping_kwargs)
|██------------------| 14.0% - epoch_loss: 0.0647465525 - val_loss: 0.0618337077
the GPU usage is minimal, yet CPU usage is high, so it's def not using the mps
:
CPU Power: 30096 mW
GPU Power: 66 mW
ANE Power: 0 mW
Combined Power (CPU + GPU + ANE): 30162 mW
Are scgen
and other tools from the scArches toolkit limited to cuda
for GPU calculations? Just to know, if it's even worth trying to make it work or if I just need to never ever run it on a mac and always go for linux if I need to train a model
When trying to install scArches as per instructions in https://docs.scarches.org/en/latest/installation.html, both listed installation options fail on arm64 osx platform.
the installation finishes correctly, but when importin
scPoli
in the following code snippet inside of a jupyter .ipynb file:it throws an error #211.
LibMambaUnsatisfiableError: Encountered problems while solving:
Could not solve for environment specs The following package could not be installed └─ pytorch-cuda 11.8* is not installable because it requires └─ cuda 11.8. , which does not exist (perhaps a missing channel).
EnvironmentNameNotFound: Could not find conda environment: scarches You can list all discoverable environments with
conda info --envs
.