My understanding is that the current moveTrappedPoints function only moves the point & relaxes the map if the new position has a lower stress than the original position. This can miss points which should be moved, but are not lower stress when moved in the current configuration (i.e. moving them and relaxing the map radically changes the map, and this new map has lower stress).
I think Derek's original implementation in lispmds was to move a point to a few new positions (maybe 10, randomly distributed in a grid), and then relax the map. This can sometimes find a better optimum. Of course, it is still limited by moving a single point, but is a useful diagnostic.
My understanding is that the current moveTrappedPoints function only moves the point & relaxes the map if the new position has a lower stress than the original position. This can miss points which should be moved, but are not lower stress when moved in the current configuration (i.e. moving them and relaxing the map radically changes the map, and this new map has lower stress).
I think Derek's original implementation in lispmds was to move a point to a few new positions (maybe 10, randomly distributed in a grid), and then relax the map. This can sometimes find a better optimum. Of course, it is still limited by moving a single point, but is a useful diagnostic.