Improve IPU convergence criteria (WMAPE)
Currently, if the factors between successive iterations vary less than a certain tolerance or if a maximum number of iterations has been reached, we consider the IPU algorithm to have converged.
Instead, we could require the (weighted) mean absolute (percentage) error between the target totals and fitted totals to be lower than a certain threshold. We would weight it to be sure that the categories with more observations fit better than those with less observations. We would probably make this threshold stricter for the groups (i.e. households) than the individuals (i.e. persons).