Could you train the ML algorithm to select the best traditional algorithm based on the distribution of the items being sorted?!