For each shape parametrisation, also do a "shape removed" model prediction. When every shapes predicted ROF changes by less than 0.1% then sufficient accuracy has been achieved to stop the iterations.
Could also potentially ease up on "n". At the moment n=5, however, with this boundary definition n=3 might be sufficient, and it would be significantly faster.
For each shape parametrisation, also do a "shape removed" model prediction. When every shapes predicted ROF changes by less than 0.1% then sufficient accuracy has been achieved to stop the iterations.
Could also potentially ease up on "n". At the moment n=5, however, with this boundary definition n=3 might be sufficient, and it would be significantly faster.