New variant of Roulette-Wheel Selection

[mmlion]

The Roulette-Wheel Selection algorithm has a problem: if the difference (in scores) between the best and the worst cars is small then probabilities of all cars are almost identical. For example, if scores are

100.1, 100.2, ... ,101.9, 102

then it is more reasonably to choose the best car with much greater probability then the worst car. But the probabilities are

4.95%, 4.96%, ..., 5.05%

I was running evolution in Sisyphus and after 30 generations evolution stops. I think it is because of wrong probabilities.

I have a simple idea to solve the problem: it is sufficient to subtract the smallest score from all scores before calculate probabilities. For the previous example after subtraction:

0, 0.1, ..., 1.9

So probabilities are

0%, 0.5%, 1%, ..., 10%