The bot must play all aspects of the game - map, fights, choices, shops etc. The same as a top level human player.
Seeds must be random with no hindsight and no quit and retry
Bot can choose any character or only play as one of them
Sample size must be at least fifty runs
This must be a normal run with no gameplay affecting mods, no exploits and the latest patch.
Related questions
I think this bot already counts, if their claims are correct: https://youtu.be/GdhHDRVigCk?si=51P__jq6Vr8oIib2
@Ernie their win rate is near our requirement but they haven't shared code. The old "we'll release it after we clean it up" trick. Too bad.
JK. The guy uses an AI to understand existing STS playing code, but doesn't actually train the model any more.
The bot must play all aspects of the game - map, fights, choices, shops etc. The same as a top level human player.
In what sense should it be the same as a top level human player? Just in the sense that it's making all the decisions? That's also true of bottom level human players.
Any ascension level? Or A20?
Heart Kill Required?
@jskf basic game so a0 no heart
But it has to be "real", no picking easy seeds or manipulation of the rng etc
@jskf yeah! there are some partial starts online in the sts discord, and useful fast libraries. i.e. independent full game simulators so you don't have to hook into the actual java sts runtime to "run" your tests.
I think 10% is a pretty low bar!
I have thought a lot about how to incentivize this, maybe with bounties, prizes, etc. but nothing has really happened yet.
@jskf Not sure this will matter, but is A0 a requirement? Or is any ascension level allowed? A1 might be beneficial.