Re: Fitness Functions

Don't use the Standard Squared Error function. As ranges increase it also increases the error, SSE takes 1/2 of the Squared Sum of the Target^2 - Actual{0..n}^2. As a consequence 1/2 of infinity does not give you a better estimate of the error. A more accurate measure is to take the Standard Squared Deviation Error or sqrt(1/n*Sum), Sum += Target^2 - Actual{0..n-1}^2.

Example:
Run 1: Actual Set Range=10 Sum=50 SSE=25 SSDE=2.23
Run 2: Actual Set Range=30 Sum=100 SSE=50 SSDE=1.82

The SSE is also not a good choice for comparing different algorithms. The SSE does not scale to the testing range, for example if two algorithms have an SSE of twenty-five and fifty after ten iterations, this indicates the first algorithm is the better choice. In addition if we ran both algorithms for thirty iterations with resulting SSE of three-hundred for the first and fifty-two for the second algorithm, our second algorithm would be a better choice. The SSDE would also indicate this with the second algorithm as the better choice at 1.31 compared to 3.16.

I do understand that the Standard Deviation is sqrt(1/(n-1)*(Actual{0..n-1}^2-Mean^2) so maybe using deviation is somewhat incorrect. Maybe it's better to call it the Post Root Mean Squared Error, unlike but similar to the RMSE.