IPF does not guarantee any convergence for the computation of the multi-way distribution. When the process Stop with a parametric number of iterations (or marginal convergence improvement delta) it returns a distribution that does not sum to one. Try to know if normalization is a good solution before implementing it ?
IPF does not guarantee any convergence for the computation of the multi-way distribution. When the process Stop with a parametric number of iterations (or marginal convergence improvement delta) it returns a distribution that does not sum to one. Try to know if normalization is a good solution before implementing it ?