Running Simba as singularity container on HPC cluster

[sgoldenlab]

Running Simba as singularity container on HPC cluster

[sgoldenlab]

Hi @singhsaluja - don't have a solution to this situation, but, some of the error messages arefamiliar.

First, we have written SimBA for Win only at this point and I do not know how it behaves in Linux.

The numpy AttributeError: make sure you are running numpy==1.18.1 and tensorflow-gpu==1.14.1 - if you already are - then this has helped me in different situations (non-SimBA related), for the same error msg, in the past: aleju/imgaug#537 (comment), I'd stick the np.random.bit_generator = np.random._bit_generator after all imports in SimBA.py

For the tk error, I have not encountered this, it seems to be related to import of matplotlib: https://stackoverflow.com/questions/55811545/importerror-cannot-load-backend-tkagg-which-requires-the-tk-interactive-fra

Matplotlib is used in the following script and imported on the following lines:
dpk_script\train_model.py: 7
gantt.py: 2
labelling_aggression.py: 8
line_plot.py: 4
plot_heatmap.py: 3
prob_graph.py: 4
sklearn_DLC_RF_train_model.py: 17
train_model\train_model_user_defined.py: 24
train_model_2.py: 23
train_multiple_models_from_meta.py: 15
validate_model_on_single_video.py: 13

I would try to adopt the matplotlib import as discussed in the thread and see if that starts it.

Also, SimBA does not require wxpython, but DeepLabCut and perhaps also DeepPoseKit does. You can try and skip wxpython installation. Try to install SimBA without tensorflow and in-build DeepLabCut/DeepPoseKit menus - pip install simba-uw-no-tf

Thanks

[MMathisLab]

To run deeplabcut on docker/singularity, i.e. headless, you must export DLClight=True before import. https://github.com/DeepLabCut/Docker4DeepLabCut2.0