sebhoerl
commented
4 months ago - Before, IPU nearly converged, but not perfectly, because the weighting totals were not consistent. Since the RP data does never have all zero-year old persons, the upscaling would be very high for the projections. That's why we excluded "zero years" from the weighting targets. This, however, should also then affect the overall population total (excluding zero-year olds) which was not the case before.
- Now, the weighting is properly implemented, excluding the stratum of zero-year old people from the weighting process, as well as those that are 105+. IPU now converges by convergence criteria of a scaling factor of less than 0.1% and not only by the maximum iterations.
- Furthermore, made the whole process more verbose: we print the IPU process with the current update factors and then print the min/max/mean overall update factors at the end of the process.
- IPU now must converge, reaching the maximum iteration limit is not allowed anymore.