I would like to reproduce twice the same results of a training in terms of wight matrices.
I already have my validation set fixed (not split during the training):
