Spring RTS Engine

Yep, SC:BW has a much stronger AI community, but the drawbacks are that you probably won't be able to compete, and the engine itself is closed source so there are limits.

(I initially made a SC:BW AI that competed in the first competition they held, and that counted as a undergrad class project.)

Concur with AF that BroodWar is probably a good target for academic projects, since it is stable, well-known, impact can be large, your AI wont break because you left the community for a month or two and there was a new release

Note that for Lua, there is an entire matrix library, machine learning library available, which is torch http://torch.ch/ . Can use GPUs https://github.com/torch/cutorch https://github.com/hughperkins/cltorch and create LSTM networks http://github.com/karpathy/char-rnn and so on. Seems like no-one has tried combining this with Spring yet, so maybe that could be an interesting direction to try possibly?

Two of those you linked wouldn't work as they require compilation unless you use something like Shard, and the third one will probably require modifications as every non-Spring library does.

Good luck generating enough data to train any deep billion-parameter (C/R)NN to reasonable levels, too.

gajop wrote:Two of those you linked wouldn't work as they require compilation unless you use something like Shard, and the third one will probably require modifications as every non-Spring library does.

Ah. Cannot call into 'c' code from the lua AIs?

hughperkins wrote:Ah. Cannot call into 'c' code from the lua AIs?

No, that's not it. Those libraries require C code to be compiled, which you can't do if you're making an AI as gadgets.

Ah ok. So would need to use it from a normal, non-Lua, AI in fact, which could choose to run a lua script, but wouldnt be a lua script in the sense of being part of the mod and so on?

You don't have to train for billions of parameters, just take a standard AI and apply the concept to a small facet, such as navigating scout planes, detecting groups of units, predicting were units are going to go, etc