FIND ME ON

GitHub

LinkedIn

Boltzmann Selection

🌱

MachineLearning

Definition

️️️️️️️In this selection method we define the notion of temperature TT as a tuneable parameter for determining selection pressure. This notion of selection pressure is the extent to which high-fitness individuals are preferred over low-fitness ones. We can implement it similarly to the two previous ones:

Roulette-Wheel Selection

The probability pip_{i} of selecting individual ii is pi=eFiTj=1NeFiTp_{i}=\frac{e^{\frac{F_{i}}{T}}}{\sum\limits_{j=1}^{N}e^{\frac{F_{i}}{T}}}Individuals are selected with approximately equal probabilities if TT is large and vice versa for small TT (high-fitness preferred).

Tournament Selection

Pick two individuals kk and jj randomly from the population. During selection a random number rr is generated and the selected individual ii is determined according to i={j\mboxifb(Fj,Fk)>rk\mboxotherwise  b(Fj,Fk)=11+e1T(1Fj1Fk)i=\begin{cases}j&\mbox{if }b(F_{j},F_{k})>r \\ k&\mbox{otherwise} \end{cases} \ \ b(F_{j},F_{k})=\frac{1}{1+e^{\frac{1}{T}(\frac{1}{F_{j}}- \frac{1}{F_{k}})}} with FjF_{j} and FkF_{k} as the fitness values of the two individuals. For large TT the probability of either getting chosen is equal, if TT is small, the better of the two is selected more often.