Discrete Optimization Algorithms
How to make neighborhoods and how to evaluate solutions are important.
SA: The temperature has role to control probability function. At first of
program execution, it should be enough high to be able to leave a local
minimum. At last of execution, it should be enough low to spend much
time to search around the global minimum.
In a simulated annealing, there are basically two probability functions:
The most important thing in MM is observing the problem
and making good problem-specific-heuristics.
I think we should learn many nice
instances of greedy algorithm, rather than many meta-heuristics.