Will you consider writing an article about your Starcraft 2 AI? I am sooo curious. (Or just send me to some additional resources?) I used to play all the time and was reasonably competitive (Diamond League).
Does your AI have APM limitations? Otherwise, I know things can get crazy.
Does it play a specific race?
Does it use a general strategy? Or will it sometimes rush, focus on macro, drops, etc?
Just super curious 😇
who might find this interesting, as well!
And I'd love to post here but it's a little intimidating, especially considering that I'm a total novice when it comes to AI. Maybe I could post about how easy it is to get started!
This is really fascinating. The idea of training your own AI to battle your opponent's version is really fun and interesting. For maps with only one spawn location, I would think a super-early proxy gateway strategy to train pairs of Zealots would be pretty deadly, especially with perfectly-managed usage of the chrono boost. For your coworker's AI to account for that threat, it might have to devote more resources for early defense and/or scouting.
I'll check out the article you linked, could be a fun weekend project (and justifiable escape back to SC2) once I get some free time.
A "how to get started" post would be great if you feel up to it; but of course, no pressure. I'm just sure a lot of us would be interested ! :)
Awesome! I've only just got my feet on the ground, here, and I've been following this guy Steven Brown's guides on Medium: chatbotslife.com/building-a-basic-...
His guides work beautifully one after another to get you started on first making a bot that can do ANYthing at all, then working up to adding reinforcement learning and saved training data, etc. Highly recommended for a great weekend project (especially since SC2 is free now!)
The AI I'm working on right now is solely Protoss, and has no limit on APM, because I'm trying to see how strong I can get it at the moment. I have a coworker who's also gotten into it at the same time as me and we battle our AIs against one another, so APM isn't an issue. As of my current level, it rushes the hell out of opponents every game, but once I tackle the challenge of ai-friendly minor unit control and placement the AI might be able to teach itself how to macro better.
That's really the most interesting thing about this entire experiment is that you really don't determine what specific STRATEGY the AI picks so much as give it a realm of options and a reward algorithm and then turn it loose for a while.
We're a place where coders share, stay up-to-date and grow their careers.
We strive for transparency and don't collect excess data.