Fix overflow dual annealing

[benvanwerkhoven]

Fixes the warning as mentioned in #362.

Also solves a possible division by zero in simulated annealing.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[stijnh]

Looks good to me!

[stijnh]

The line (np.abs(old_cost) if old_cost != 0.0 else 1e-20) could also be max(np.abs(old_cost), 1e-20) to make sure it also works if 0 < np.abs(old_cost) < 1e-20

[benvanwerkhoven]

That line is only to avoid a division by zero, so the 1e-20 is only used when old_cost is exactly 0.0. If the tuning objective is meaningfully trying to optimize tiny values and 1e-20 would be 'higher' than the real objective values this could indeed lead to strange decision making. However, having actual objective values in that range with Kernel Tuner seems unlikely and it would still be extremely rare as this is only triggered when the returned cost is exactly 0.0.