Train on with other seeds

[sandrinebedard]

Currently, our mdoel is only trained for one seed , 15, would be good to test other seeds.

[naga-karthik]

Yeah, this has to be done but we can't compare our results with ivadomed anymore if we change seeds. We can only do a comparison between nnUNet and MONAI (which could be a comparison between hard and soft labels).

If we are positive about removing ivadomed in our STD CSA and per contrast CSA plots, then I can starting thinking about training on different seeds with nnunet and monai

[sandrinebedard]

Also redo a STD per subject for each seed to get more subjects in the violin plot

[naga-karthik]

Just a quick note of clarification -- choosing different seeds might result in a slight overlap of test subjects coming from each seed. While we will surely have more points in the STD CSA plot, it will simply not be num_seeds x num_test_subjects (i.e. we would have to remove some repeated subjects)