4 ms·
Yeah, I did the same (VanillaJS) a couple of years back, and started thinking about implementing a computer opponent, but it's really not clear what the evaluat
by mrec 8y ago
Yeah, I did the same (VanillaJS) a couple of years back, and started thinking about implementing a computer opponent, but it's really not clear what the evaluation function should be.
- formalsystem 8y agoYou need only codify the victory and loss conditions to build an AI using standard Reinforcement learning algorithms. Feel free to message me if you want more detail on how this would work. We should be able to hammer something out in less than a day.
- mrec 8y agoThanks for the pointer, but I'm afraid I have zero knowledge of ML and wouldn't be any help. I have to say I'm a bit sceptical that this approach can work purely based on win/loss conditions - an attacker victory in particular can often take hundreds of moves in Tafl games.
- DrHuman 8y agoAgreed. The only thing I could think to do would be look for pieces to capture in any given turn, but that would never get it to an end game condition. It's beyond my current capabilities to make a proper computer opponent, but it would be a good exercise.
- mrec 8y ago> look for pieces to capture in any given turn Not as a general rule. If you're playing as the attacker, you don't always want to take defender pieces even if they're "free" - they're more useful boxing their own king in.
- DrHuman 8y agoYea, it's not even a good strategy really. Just the only thing that I felt like I could code that would be a step up from just making a random legal move.