Add documentation for creating Conda environment "no_cuda" / older Macs

[pollytur]

Hello!
I tried to install sleep on Mac (Intel i7, Monterey 12.2.1, no GPU) via Conda but it failed

$ conda env create -f environment.yml -n sleap
Collecting package metadata (repodata.json): done
Solving environment: failed

ResolvePackageNotFound: 
  - cudatoolkit=11.3.1
  - nvidia::cuda-nvcc=11.3
  - cudnn=8.2.1

I used git clone https://github.com/murthylab/sleap && cd sleap and conda env create -f environment.yml -n sleap before as conda create -y -n sleap -c sleap -c nvidia -c conda-forge sleap=1.2.2 does not work for mac.
Seems like, indeed, there are no these packages for a laptop without GPU. I wanted to landmark the videos on my usual laptop and then to train on a server with a GPU. Could you please let me know if there is such an option? And if yes, how should I solve the thing above? (I already tried conda install -c conda-forge <package_name=version> and it did not help)

Thanks in advance!
Polly

[pollytur]

I also tried to run the following

1. conda activate sleap_trial
2. pip3 install sleap
3. sleap-label

But it failed with the following:

(sleap_trial) $ sleap-label
Traceback (most recent call last):
  File "/usr/local/bin/sleap-label", line 5, in <module>
    from sleap.gui.app import main
  File "/usr/local/lib/python3.9/site-packages/sleap/__init__.py", line 9, in <module>
    from sleap.io.dataset import Labels, load_file
  File "/usr/local/lib/python3.9/site-packages/sleap/io/dataset.py", line 80, in <module>
    from sleap.gui.dialogs.missingfiles import MissingFilesDialog
  File "/usr/local/lib/python3.9/site-packages/sleap/gui/dialogs/missingfiles.py", line 154, in <module>
    class MissingFileTable(QtWidgets.QTableView):
TypeError: 'Shiboken.ObjectType' object is not iterable

[roomrys]

Hi @pollytur,

Thanks for bringing this to our attention. In response to the conda from source installation error, since you are not training on your own computer, we can try removing the CUDA libraries from the dependencies. Can you try editing the environment.yml file in the main (outer) sleap directory. Remove or comment out the following lines under dependencies:

    - cudatoolkit=11.3.1
    - cudnn=8.2.1
    - nvidia::cuda-nvcc=11.3

Then, from within the main (outer) sleap directory, run the conda env create -f environment.yml -n sleap command again. I am trying it out as well on my end to ensure that nothing else breaks when we do this (however I am on Windows). Let me know how that works for you.

Thanks,
Liezl

[pollytur]

Thank you!
Seems like this works. I will let you know if I will find any suspicious behaviour.

Thank you one more time,
Polly

[roomrys]

Perfect! We will work on a more permanent fix, but glad to get you up and running.

[roomrys]

Another option is to create the environment using the environment_no_cuda.yml. However, we have no documentation on this.

[roomrys]

I added a note in the conda from source installation where we specify to use the environment_no_cuda.yml if the user's computer does not have a NVIDIA GPU.