I have trained a keras model, with the (apparently) best weight giving an accuracy for the trainset of 0.9966 and loss of 0.01087 and for the testset 0.9867 and 0.09745 resp
