So I’ve actually had the next set of 5 rounds ready for about a week, but I haven’t written about them yet or proceeded with further rounds for a couple of reasons.
First, I want to re-write the code where I play against the AI so that it gives a nicer output. I haven’t done this yet, so I haven’t played against the AI yet.
Second, some of the models are actually approaching a 50% win ratio against the smart random player. I’m concerned about the quality of the training data going forward, and I know I’ll eventually have to rework how I generate the training data so that it’s useful for the models to keep learning and improving from.
In the interest of expediency, however, I decided to just punt on both those issues.
I’ll not play against one of the models again until I re-write the display code, and I’ll just keep running rounds as I am right now. If the models stagnate and fail to improve after a point, I’ll see it happen. If I’m wrong and they keep improving, then, hey, that’s even better.
Hopefully I’ll have things ready to show by round 30, but if not I’ll just keep going to 35 or even 40 before I post about it again.