Improve reproducibility across training/prediction runs

[eridona]

Hello, I noticed that when runnning training/prediction on the same dataset (same training set, same prediction set) multiple times I get significantly different results. This happens both when running on the same computer (i.e. no changes in configuration) as well as when running on different computers (same operating system, same version, so I suppose most of the dependencies will be of the same version). I was wondering if there is a way to improve reproducibility or if I might be under/overtraining. Thank you for your support! Erika