When getting the list of valid algorithms, the CLI doesn't check if the algorithms are subclasses of some other GA. This causes some algorithms to appear which aren't actually runnable (e.g. CellularGA).

Should be a simple fix.

It would be really helpful to have some progress messages logged to stdout while experiments are running. Ideas:

• time elapsed
• generation counter
• trial counter

Since GAs are stochastic, we need to run each experiment several times and collect aggregated stats on the whole run. Current results are reported for only a single run.

Maybe add some module or CLI action to run a proper experiment?

Currently, the replacement strategy replaces one of the 5 worst members of the population.
I'm worried that this is too aggressive. Takeover time seems low and based on the observed rate of fitness improvement, it seems that the lack of diversity maintenance is causing trouble.

I'd like to switch to an inverse roulette-wheel selection (where fitter individuals have a lower probability of being selected). Any thoughts @Mohamad9922?

Need to add GAs that use a dynamic mutation rate and self-adaptive mutation and selection parameters.