Starcraft AI blog

The AIIDE registration list is available today. There are 14 bots on the list, compared to 15 last year. I count 7 protoss, 5 zerg, 1 terran, and 1 random, though among the updated returning bots protoss and zerg are 3-4. Zerg seems to be gaining in popularity. (As it should, ahem. I have been ahead of the curve the whole time!)

The familiar bots, in order of their BASIL ranks today:

Protoss dominance is showing, but it already cracked in CoG without the help of Monster. Terran shyness is also showing, but I notice that WillyT has improved a lot in the last year. I counted new zerg Fresh Meat as a familiar bot even though neither Fresh Meat nor its terran counterpart Halo by Hao Pan has participated in AIIDE before—Hao Pan is an old stalwart. It will be interesting to see how Fresh Meat performs in the different world of a long tournament. Steamhammer is last on this list, but I have already fixed key weaknesses. It will perform better in the tournament.

We have 3 newcomers, a fair number. Last year there were 4 newcomers, and unfortunately only EggBot ended up playing.

Eujain Ting seems to have some experience. I found old repos related to Broodwar and BWAPI on Bitbucket: Eujain Ting repositories. And I see Eujain Ting registered for the 2011(!) AIIDE tournament without playing. The name “Blue Soup” (after the debris of a destroyed dragoon) suggests either low expectations or a sense of humor! Kim TaeYoung last year registered protoss DanDanBot (which did not play). This year, the zerg name “real5drone” suggests a 5 pool strategy, whether honestly or otherwise. Wang Bin last year registered Taij, which I’m guessing was a typo for this year’s allusive name Taiji (and which also did not play). From past experience I do not have high hopes for the newcomers, but occasionally a great one appears. And I think it’s valuable experience to participate no matter how good or bad your bot is.

Plus 3 holdovers from previous tournaments, with their win rates from last year.

It’s too bad that we don’t get an updated Dragon. It’s a complicated and interesting opponent. UAlbertaBot is here for yet another year, I gather, mainly as a long-term baseline to measure progress. I forecast that it will score less than 15% against the updated bots, and this year or next it is likely to lose its value as a baseline. 2017 champion ZZZKBot scored 39.89% last year and is not being carried over—now that’s a sign of progress.