In my program, a creature is an 8-game 6-table setup. A combination of two creatures uses a few games from each setup. A mutation swaps two seating positions within a game. These definitions were arrived at with almost no thought but upon running it, I came up with a solution which had 31 bad seatings in about one day! This was incredible but it was difficult to reproduce, difficult to improve on, and not much better than 33.
While bragging to my mom about these amazing results, she mentioned that the bridge players didn't like being "player 19" on the current seating arrangement because he sits at the same table too much. She also pointed out that having the same partner twice was a major no-no.
The solutions that I had created with my Genetic Algorithm scored 31 and 32 but were pretty bad in terms of bad partnerings and what I have come to call "bad tablings" -- repeatedly sitting at the same table.
To try to eliminate these problems, I modified my scoring algorithm to score 5 points for the bad seating, 10 for bad partnering, and 2 for bad tabling.
Starting with completely random creatures, and continuously mutating through several million generations I have not yet been able to do significantly better than the setup with 33 bad seatings.
After many iterations of tweaking the algorithm and restarting
with a new set of random creatures I came up with a great idea.
I could spike the soup.