CIFAR10 FEDAVG SIMULATION ERROR

[FabioNotaro2001]

Hi', I'm trying to execute the example of cifar10_fedavg present in this link: https://github.com/NVIDIA/NVFlare/blob/main/examples/cifar10/cifar10-sim/README.md

In particular, I put myself into the directory cifar10/cifar10-sim and I executed the commands
python3 ../pt/utils/cifar10_download_data.py
export RESULT_ROOT=/tmp/nvflare/sim_cifar10
./set_alpha.sh cifar10_fedavg 0.5
nvflare simulator job_configs/cifar10_fedavg --workspace ${RESULT_ROOT}/fedavg_alpha0.5 --threads 8 --n_clients 8

But after this last command, I expected the training to start, instead I got this long sequence of errors:

[YuanTingHsieh]

Hello, thanks for the interest.

The problem is "No module named 'pt'".
Did you do "export PYTHONPATH=${PWD}/.."?

Can you try to check if your PYTHONPATH include its parent?

[FabioNotaro2001]

Yes , now it works! Thank you very much. One last question: why this command isn't inserted in the simulator example with hello-pt-tb and instead here it is necessary?

[YuanTingHsieh]

good question, we will try to standardize example structure in the future.

But now they are a bit different.

So if you compare the configurations between these two examples, you will find that cifar10 needs codes from its "pt" folder.
Which is this folder: https://github.com/NVIDIA/NVFlare/tree/main/examples/cifar10/pt
So that's why it requires to add this path.
The reason it has this "pt" folder outside is because it wants to reuse the same code for all cifar10 jobs.

VS

hello-pt-tb example we only use things in "custom" folder, and NVFlare add codes in "custom" folder automatically.