Like the Minesweeper post, this won’t be nearly as fun as the “This was my process” post. Instead, it’s just a postmortem on how I thought it all went.
This time I didn’t do as well as I’d hoped. It’s hard to rank the AIs, because we did a double elimination bracket system, so I didn’t get to play against everyone, but there were a few consistent factors.
My AI was roughly at the higher end of the middle of the pack. I could beat most AIs, but a few people had AIs who were consistently better. I actually lost to the same person in both rounds of my losses in very close matches that could have gone either way. In the round where I got knocked out, if I had won, I would have beaten him, it was that close.
There were some AIs that played less conservatively with their cards and would remove high cards from their hand very fast. In cases where they would get Gin, my AI might be sitting on a few dozen points. Needless to say, those AIs generally beat mine.
Compared to AIs that had a similar strategy to mine, it did generally well. A few had a similar strategy to my own, doing things like removing pairs of Kings and keeping single cards of low cost.
Alan once again was the definitive winner. His AI never lost a single match against anyone else (for clarification, a match was 40 games). I asked what he did, and he used a similar base as my own, but improved it with weights on discarding certain cards. As he fine tuned the weights, the AI got better.
I do wish I had added that myself, but time was unfortunately sparse, the capstone project consuming the majority of my time.
Either way, the next project is a custom made Tank Game, which is thoroughly complex (to code, it’s fairly simple to actually play), so that will be an interesting challenge. Because it’s custom made, right now it’s got a lot of bugs. Certainly going to be interesting with all the fixes coming in throughout. Until then.