Tic-tac-toe with Charlie

I first attempted the Tic-tac-toe experiment, using Charlie, with the same setup as the symbolic regression experiment, along with a MakeMove function node and terminal nodes to represent the status of each square on the board:

Function nodes	Add, Subtract, Multiply, Divide, MakeMove
Terminal nodes	0, 1, 2, EphemeralConstanct(0,10), SQ_00, SQ_01, SQ_02, SQ_10, SQ_11, SQ_12, SQ_20, SQ_21, SQ_22
Crossover probability	90%
Reproduction probability	9%
Mutation probability	1%
Mutation depth	2
Max tree size (number of nodes)	unlimited
Max tree depth	unlimited

That didn't produce very good results. This was as expected as there were no conditional function nodes. So I introduced If-Then-Else, Equals, Greater-Than, Less-Than, And and Or operators and immediately got more promising results.

However, it was still very easy to beat the evolved opponents, so I went ahead and made various performance optimisations including making the simulation multi-threaded. Still not very good results.

I then tried running the setup with huge populations, on a 64-bit OS with lots of RAM, for several weeks but still the best opponents were mediocre and very easy for humans to beat. They frequently made large blunders and you didn't have to be particularly careful to ensure a win against them.

All in all, very similar results to the initial Beagle experiment, so quite disappointing. That brings my journal up to the present. Yes, I'm stuck with this current issue of the simulation getting stuck before strong players are evolved. I'm going to try and think of some new approach or tweak to the setup to improve things. Do let me know if you have any suggestions!