We try to provide the main python packages, specially the once that needs optimizations for our system. If you need additional python packages that can be easily installed with pip, then we recommend that you install this in your home area. If you install this in one of the ML nodes then it will be available to be used all the other nodes.
For example if you need graph_nets https://github.com/deepmind/graph_nets, we have installed the basic dependencies and you could follow the procedure below to get it installed din your home area.
[sabryr@ml7 ~]$ module purge [sabryr@ml7 ~]$ module load TensorFlow/2.4.1-fosscuda-2020b [sabryr@ml7 ~]$ pip install --user graph_nets
- If you face any issue check the following first and send us the information when you request for support
# Check where it is loading from
[sabryr@ml7 ~]$ python -c "import graph_nets as gn; print(gn.__file__)" #For the user sabryr it will be /itf-fi-ml/home/sabryr/.local/lib/python3.8/site-packages/graph_nets/__init__.py If you
# If you get "undefined symbol: " error this could mean that you had already installed some dependencies that the Tensorflow module provides and there is a conflict. To resolve this you need provide a different instalation locition than the default or clean install on .local
mv $HOME/.local $HOME/.local_backup
Then install again following the above procedure.