- Added function checkprops, set default tolerance parameter to 1.0e-4
- Using "area" instead of "square".
- Options "-equal\notequal" isn't used together with "-s" in blend and offset test cases.
- Correct regressions/differences/improvements and CPU problem (set props tolerance to 0.1)
- Corrected test cases to use checkprops proc.
- Correct image difference
- Updated TODOs in test cases.
- Updated test cases to get correct images of result shape
The patch changes the algorithm of choosing the initial approximation for Newton's method. Instead of searching for a point on the fine shifted grid, the algorithm performs initial search for candidate points in the original coarse grid (which is cached in new version). After that particle swarm optimization (PSO) is used to localize the global minimum. This algorithm optimizes a problem by having a population of candidate solutions ("particles"), and moving these particles around in the search-space according to simple mathematical formula over the particle's position and velocity. Each particle's movement is influenced by its local best known position but, is also guided toward the best known positions in the search-space, which are updated as better positions are found by other particles. This strategy has reported good results in solving complex global optimization problems.
Current patch implements initial version of PSO for using in Extrema_GenExtCS class. Typically new approach allows to reduce the number of evaluations by 5-10 times. There are only few cases there the numbers of evaluations are comparable.
