Learning in microRTS bots for the IEEE-GIC 2018 competition

[Santiago Ontanon]

Hi, I have received several requests to allow some form of learning in microRTS bots. Since I did not want to change the rules too much just a few months before the deadline, I decided for this intermediate solution:

- The very first time that a bot plays in a given map, the bot will be given the map, and 1 hour of time to run any learning process they want in such map.

- After that, the bot can write to disk whatever it wants (a "write folder" path will be given as a parameter)

- Each time a new game starts, the bot can read from that folder.

I'll be developing the API for this in the next few days. But the current API already allows for something like this via the "preGameAnalysis" call. The only thing that will be added is the possibility to write to disk after this analysis. I'll be releasing more details in a few days, once I have the API figured out. But any feedback is welcome.

The goal of the competition is just to help people out with their research by allowing comparison across approaches. So, if there is any additional such feature that would be required by some of your approaches and that the competition setting of microRTS does not support, by all means let me know :)

[Santiago Ontanon]

Ok, this is implemented. The first time an AI plays in a map in a tournament, the "preGameAnalysis" call will be called with 1 hour of time, so that you can do any learning you want on that map (you will NOT know who is your opponent, you will NOT know the name of the map, and you will NOT know if you play as player 1 or player 2). You can save anything to disk in the provided readWriteFolder. Subsequent times you play in the same map, this function will be called again, but just with 1 second of time (this is just so that you have time to read from disk anything you had saved).

Any suggestions/comments are welcome! :)