When we provide a test size that is lower than the train size, the model will try to pull some images that do not exists and the training will crash. See here.
When we remove the test visualization, the code will still try to plot the test metrics, leading to another crash.
The first issue is still problematic, my fix will only remove the crash but it does not modify the current behaviour that will always try to visualize the same amount of images, whether we are visualizing the test or train images. It could be fix in another PR.
This PR fixes two issues :
The first issue is still problematic, my fix will only remove the crash but it does not modify the current behaviour that will always try to visualize the same amount of images, whether we are visualizing the test or train images. It could be fix in another PR.