Initial Objective Scores

I completed the objective scoring function that I mentioned earlier and wanted to document some initial results. I did 10 evolution runs against each opponent type, until a perfect fitness case was found, and calculated the average objective score:

Random Move: 124, 180, 108, 132, 366, 160, 108, 72, 198, 108 =>155.6
Good Move: 310, 344, 882, 138, 273, 1172, 150, 518, 128, 246 => 416.1

The Random Move opponent simply just picks a random, valid, move each turn.

Good Move picks a winning move if there is one move that makes 3 in a row, and blocks the opponent if there is a chance for them to play one piece to get 3 in a row. Otherwise it just picks a random, valid, move.

The setup was as follows:

Function nodes	+, -, *, /, IfThenElse, =, >, <, And, Or, MakeMove
Terminal nodes	0, 1, 2, Ephemeral Constant (0-10), SQ_00...SQ_22
Crossover probability	90%
Reproduction probability	9%
Mutation probability	1%
Mutation depth	2
Max tree size (number of nodes)	100
Max tree depth	30

Of course, this was more teaching the AI how to play as opposed to letting it evolve from scratch, so the true test would be with co-evolution. However, I needed to do some more groundwork to get that up and running in my new environment...