Re:Solving Problems with MiniZinc
by CJefferson on Monday October 23, 2017 @06:03AM
(#15531982)
Reordering constraints changing performance is basically impossible to avoid for constraint systems, because we are very bad at predicting which constraints should be considered first -- well, you can always apply some normalisation pass which at least hides the problem from users.
Reording causing changes in feasibility of a problem is a MAJOR bug, and I hoped they reported it. The constraint system I work on (Minion and Essence', a competitor to MiniZinc), has (AFAIK) never had such a bug in any release (they come up occasionally during development, but we have serious tests specially designed to detect that kind of problem).
For constraints x>=0 and x=0, switch the order for a performance improvement. This shows solvers have preferences. MZ chooses the solver based on reasons.
MZ seems to work beautifully for simple/modular examples. A professor suggested this one instead (for performance, popularity): http://eclipseclp.org but I can't yet assert if they are comparable or if the statement is accurate.
From what I've seen I had the impression that MiniZinc had some problems, i.e. it has some unexpected behavior, such as: - reordering constraints changes performance - reordering constraints may change results/feasibility of a problem
For quick prototyping it was nice though.