You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Trying to get a two-site swarm training on the Duke dataset to run, I noticed the following:
The jefftud/nvflare-pt-dev:3dcnn (a3bd4e8a9577) image does not work for me, I think there are two issues: a psutil version problem as Steffen described in July, and the PYTHONPATH variable that should be set to /workspace/controller/controller. I've continued with a locally built image from docker_config/Dockerfile_3dcnn from the dev-7-test-controller branch (planned to be merged to main, #7). The image should be updated accordingly.
Moreover, simulation mode does not work out of the box, but requires having a dataset available and split somehow (or just symlinked to the full folder) matching the simulated node names. This should be explained and/or included in the scripts.
Trying to get a two-site swarm training on the Duke dataset to run, I noticed the following:
The jefftud/nvflare-pt-dev:3dcnn (a3bd4e8a9577) image does not work for me, I think there are two issues: a psutil version problem as Steffen described in July, and the PYTHONPATH variable that should be set to /workspace/controller/controller. I've continued with a locally built image from docker_config/Dockerfile_3dcnn from the dev-7-test-controller branch (planned to be merged to main, #7). The image should be updated accordingly.
Moreover, simulation mode does not work out of the box, but requires having a dataset available and split somehow (or just symlinked to the full folder) matching the simulated node names. This should be explained and/or included in the scripts.
This might resolve part of #2.
AC
The text was updated successfully, but these errors were encountered: