Starcraft AI blog

Yesterday’s loser’s round match KrasioP-Steamhammer ended up with the result I expected, but not in the way I expected. Krasi0P has two builds, cannons in base and proxy cannons. Steamhammer can easily beat either one if it correctly predicts it, but it is poor at predicting. The wins tend to go in streaks, Steamhammer loses a bunch in a row then wins a bunch in a row, because both bots are somewhat slow at reacting when the opponent’s build switches. And Steamhammer’s recent games against Krasi0P, the 2 games in the SSCAIT round robin, were losses. So most likely the losing streak would continue.

1-1 after the first 2 games was good. In the last game, my initial diagnosis was the same as the commentator’s: Steamhammer was going to play 1 hatch muta, an excellent choice that should win. I was suddenly looking forward to a surprise match victory. But when I saw the second hatchery start, I knew something had gone wrong. Steamhammer ended up playing a nonsense build that countered nothing, a sure loss. After seeing the cannons it should have turned a few planned zergling pairs into drones; that part looked OK. I’m not sure what failed, but my best guess is that something caused Steamhammer to break out of its opening build and fall back on the strategy boss, which is buried up to its neck in ZvP weaknesses (it’s Steamhammer’s weakest matchup). Those mistakes look like strategy boss mistakes. But nothing I saw in the game should have caused it to break out of the opening. It may be a bug I haven’t seen before.

In any case, the fixes in version 3.3.6 made Steamhammer substantially stronger against Krasi0P (and many other opponents). Steamhammer will look impressive if it ever gets into a tournament with all the major bugs fixed before instead of after....

Really next, no really, this time I mean it: Steamhammer’s experience versus cannon rushes.