the visualization of breast density challenge tutorial using tensorboard

[GhanemBahrini01]

Hello all,

So I have run the tutorial code https://github.com/Project-MONAI/tutorials/tree/main/federated_learning/breast_density_challenge and when visualizing the results I was wondering what is the difference between " val_local_model_acc " and " val_acc_local_model_acc " (the visualization result after running " tensorboard --logdir=./result_server")!

[holgerroth]

Thank you for your question. val_local_model_acc and val_acc_local_model_acc are computed at different phases of training. Metrics with val_local_model_* are computed after the local training loop and metrics with val_acc_local_model_* are also computed during the local training at certain intervals (controlled by val_freq).
If aggregation_epochs = 1, then both plots will be the same.

Note, val_global_model_kappa is calculated separately with each site's data at the start of each round, which is why you see 3 separate traces. If you look at the leaderboard, you will see that statistics are tracked both separately for each site (i.e. Lin. Kappa 1-3) and averaged across all sites (i.e. Lin. Kappa all).