* adding debug-model to print the internals for debugging purposes * expend debug-model script with 4 stages: before, init, train, predict * avoid enforcing to have a seed in the train script * small fixes