On 13-04-13 00:42, Alexander Brock wrote:
... 2. In most tests the method stops at some point because no improvement occurs no matter how small the step size is chosen allthough the fit isn't perfect yet. See path 2/3 for examples. Currently the algorithm just stops but I'd like to continue somehow. I could choose a random search direction and try to find a better solution or maybe just take random samples from a small neighbourhood of the current set of parameters.
If I understand correctly you roughly do the following: 1) Find points on original curve(s). 2) Estimate new curve. 3) Find correspondence between points on the original curve and the new curve. 4) Move new curve closer to original curve. 5) Repeat 3 and 4 until convergence.
This is not an uncommon set up (lookup EM optimization for example, there is also a less statistically-minded method like that, but I can't remember the name right now), but can be tricky to get right. In particular, if you consider both the the correspondence you compute and the "fit", then I doubt both steps improve (or maintain) both, and this might give you some trouble.
- make -j3 takes > 2m to finish even when I only change one line in
bezier-utils.cpp on a Core2Duo, 2,4GHz, is that normal? make install then takes >1m. I use ccache, g++ and turned optimization off.
More or less, linking often takes most time for these small compiles.
- Where can I find examples for automated tests?
Is this what you're aiming at? http://wiki.inkscape.org/wiki/index.php/Testing#Developers