Hi Joost, The training set isn't a problem to compute, and necessary to know what is the probability of obtaining any given sum. The problem is more on how to scale up the recursive convolution. FFRANK