Optimization is post hoc here : you have to train first to be able to huffman en ode, so it's not a pure format question