size of tournament vs size of population

[Florin Andrei]

I've started from a simple example, which I've expanded quite a bit with a much more complex objective function - I'm using GA for feature selection for regression models, and the objective function indicates how well the models fit the data.

https://github.com/DEAP/deap/blob/master/examples/ga/onemax.py

In this example, the population size is 300, while the tournament size is 3.

Speaking in general, if I increase the population size 10x, should the tournament size be increased proportionally? I'm aware that would be pointless in the simple example linked above, but replace the objective function with something much more sophisticated (select features, fit regression model, evaluate model performance, etc).

It seems like they should be proportional, but perhaps my intuition is wrong.

The goal is to avoid falling into local extremes.

Thanks!

[EBo]

It has been several years since I played with this, but there is a
delicate balance of population size, number of genes, mutation rate,
cross-over rate, and how quickly the population fixates on an answer.
If your population is too small, then it fixates, and you are dependent
on mutation only. If your population is LARGE, then it takes forever to
converge on a solution.

When I used DEAP before, I played around a little to find a nice
balance. It has been long enough, I do not remember how many genes and
population size, seemed to work well. I should add that my evaluation
function took 1 to 10 hours to run for each individual, so it was worth
the time it took in tuning.