https://github.com/KatherLab/MediSwarm/blob/include_challenge_teams1-5/assets/readme/README.participant.md#:~:text=Output%20files%20are%20located%20in%20the%20directory%20of%20the%20startup%20kit%3A
https://github.com/KatherLab/MediSwarm/blob/include_challenge_teams1-5/assets/readme/README.participant.md#:~:text=Output%20files%20are%20located%20in%20the%20directory%20of%20the%20startup%20kit%20(note%3A%20unlike%20local%20training%20results%2C%20this%20is%20not%20in%20the%20startup%20directory)
The instructions here dose not reflect the real places to find the training results.
Ole Schwen [下午 4:11]
I had a look again at the output of the 1.0.2-dev.260319.d7e0125 training. I only ran a preflight check, not a full local training, and obtained files startup/runs/MEVIS_1/challenge_4LME_ABMIL_unilateral_2026_03_19_162024/aggregated_model_gt_and_classprob_train.csv and site_model_gt_and_classprob_train.csv, site_model_gt_and_classprob_validation.csv, aggregated_model_gt_and_classprob_validation.csv containing output for one epoch, also for the 5Pimed and 2BCN_AIM models. Since the preflight check runs one epoch, this likely means we also get the corresponding output for local trainings. Should also work for the swarm training unless you somehow managed to change the mechanism to write these files so that it only works for the local trainings 🙂
[下午 4:14]https://github.com/KatherLab/MediSwarm/blob/main/assets/readme/README.participant.md describes where they are located for local and swarm trainings.
https://github.com/KatherLab/MediSwarm/blob/main/scripts/evaluation/README.md describes how plots can be generated from these files, I guess using one such folder structure per model.
Jeff [晚上 6:04]
Shouldn't it be under $SCRATCHDIR/runs/? （已编辑）
Debora Jutz [晚上 11:59]
For me it is under $SCRATCHDIR/runs/$SITE_NAME
Ole Schwen [上午 9:31]
For local trainings or also for swarm trainings?
Jeff [上午 9:31]
For me was checking swarm learning
Ole Schwen [上午 9:33]
Then I got the paths wrong when writing the readme. I'll create a PR to fix it
Ole Schwen [中午 11:12]
I'm confused … if I run a two-node swarm training using the MST model via the integration test, the files end up inside the startup kit directory, not in what I believe to be the scratch directory. However, that directory is suspiciously empty, there is not even runs. Maybe I got the setup wrong and something differs from the actual swarm training.
https://github.com/KatherLab/MediSwarm/blob/include_challenge_teams1-5/assets/readme/README.participant.md#:~:text=Output%20files%20are%20located%20in%20the%20directory%20of%20the%20startup%20kit%3A
https://github.com/KatherLab/MediSwarm/blob/include_challenge_teams1-5/assets/readme/README.participant.md#:~:text=Output%20files%20are%20located%20in%20the%20directory%20of%20the%20startup%20kit%20(note%3A%20unlike%20local%20training%20results%2C%20this%20is%20not%20in%20the%20startup%20directory)
The instructions here dose not reflect the real places to find the training results.
Ole Schwen [下午 4:11]
I had a look again at the output of the 1.0.2-dev.260319.d7e0125 training. I only ran a preflight check, not a full local training, and obtained files startup/runs/MEVIS_1/challenge_4LME_ABMIL_unilateral_2026_03_19_162024/aggregated_model_gt_and_classprob_train.csv and site_model_gt_and_classprob_train.csv, site_model_gt_and_classprob_validation.csv, aggregated_model_gt_and_classprob_validation.csv containing output for one epoch, also for the 5Pimed and 2BCN_AIM models. Since the preflight check runs one epoch, this likely means we also get the corresponding output for local trainings. Should also work for the swarm training unless you somehow managed to change the mechanism to write these files so that it only works for the local trainings 🙂
[下午 4:14]https://github.com/KatherLab/MediSwarm/blob/main/assets/readme/README.participant.md describes where they are located for local and swarm trainings.
https://github.com/KatherLab/MediSwarm/blob/main/scripts/evaluation/README.md describes how plots can be generated from these files, I guess using one such folder structure per model.
Jeff [晚上 6:04]
Shouldn't it be under $SCRATCHDIR/runs/? （已编辑）
Debora Jutz [晚上 11:59]
For me it is under $SCRATCHDIR/runs/$SITE_NAME
Ole Schwen [上午 9:31]
For local trainings or also for swarm trainings?
Jeff [上午 9:31]
For me was checking swarm learning
Ole Schwen [上午 9:33]
Then I got the paths wrong when writing the readme. I'll create a PR to fix it
Ole Schwen [中午 11:12]
I'm confused … if I run a two-node swarm training using the MST model via the integration test, the files end up inside the startup kit directory, not in what I believe to be the scratch directory. However, that directory is suspiciously empty, there is not even runs. Maybe I got the setup wrong and something differs from the actual swarm training.