archive by month
Skip to content

a few items

I am working furiously on Steamhammer, not leaving myself much time for posting. The version I submit for AIIDE should be substantially stronger than the current release version, with visible changes that close observers will notice.

I doubt it will have much chance versus Stardust, though. After watching games where Stardust lost, I wrote 3 new openings to try to exploit its weaknesses, to at least make a dent. One was the hydra opening that Jealous suggested in a cast. But no, a rubber ball does not make a dent in a concrete wall.

A new bot EggBot appeared on the AIIDE 2020 participant list at some point after I wrote up the participants. I guess it must have been omitted by mistake. It is by Nathan MacNeil of the Memorial University of Newfoundland (Dave Churchill’s institution), who is apparently a graduate student. A grad student should have an interesting project in mind, so I have hopes for EggBot—weak or strong, I hope it will be interesting.

I still want to post about encouraging new bot authors, but you can tell it’s not my actual priority because I’m spending my time on coding and testing instead. Still, it’s an important discussion. The addition of S A B C D E F ranks on the BASIL rankings could be a good step if it encourages new authors to make progress: “I want to move up to D!”

Trackbacks

No Trackbacks

Comments

Tully Elliston on :

Good luck with the cramming, don't introduce too many last second bugs for the tourney :-)

Jay Scott on :

Right, I’ll aim for just the correct number. :-)

MarcoDBAA on :

Stardust has some weaknesses for sure. It does not (mostly) attack expansions, and there is no real late game. Saw a RH as SH game vs Stardust, where SH should have won, but it retreated through the goons and lost most of its army. Defiler also failed to cast Dark Swarm. https://basilicum.bytekeeper.org/bots/Stardust/Stardust%20vs%20Randomhammer%20Destination%20CTR_CF9C7E21.rep

Crona and Proxy (Stardust looked a bit broken vs Proxy however) lately won one game vs Stardust.

I like the BASIL changes, especially: "Matchmaking now selects opponents with similar ELO with higher likelihood." I think, that this already improves the quality of games for everyone.

BASIL also enabled some bots, like the tscmoo bots (missed them, but terran was active before already), and other good, medium and also a few low tier bots. Good selection.

But SSCAIT still has more active low tier bots than BASIL really. The problem with win rate deactivation is of course, that it will slowly remove bots like a game of musical chairs. Disabling the weakest bots, will decrease the win rate of the slightly better bots (they lose their winning match ups), and these will be removed too, until all/most former low tier bots are depleted.

Is this the case really?
Well, see here: http://satirist.org/ai/starcraft/blog/archives/803-new-bot-RedRum.html

"At the moment, Roman Danielis has the 5th lowest elo rating [exactly true @ BASIL in 2020 too].
This clearly makes it one of the low tier bots in 2019. His bot was never updated, so we can check how well it did in earlier years. In 2014 https://sscaitournament.com/index.php?action=2014 Danielis won 26 games and lost 15 games, and would have qualified for the knockout round, if it would have been played with 16 bots at that time."

Real question should be, how weak bots are allowed to be, because win rate deactivation will not demonstrate this "objectively". If asked for I could name some older weaker bots, that play comparatively solid (expand, good unit mix or/and create interesting games, react to trouble, no crashes and slow downs, can finish all games). Roman Danielis as (nearly) easiest opponent is a bit tough I think (see above), but people might disagree of course. I personally also wouldn´t activate too many weaker bots however, just a few.

On the other hand, all bots get less games, if you have more active bots. I mostly watch top bots too ;), BUT watching new low tier bots also became more boring, because there are no bots left, that allow them to have a fun game (similar ELO matchmaking might help here however).

Tully Elliston on :

I think best solution would be to have two ladders - invite only top bots ladder + all comers ladder with anything that still works enabled.

That way you have more of a route of progression as a new bot author.

MarcoDBAA on :

There is this one already, that is mostly a top bot ladder for now (de facto):
https://www.cs.mun.ca/~z24rmk/starcraftailaddertest/home

But nah, I wouldn´t activate everything, that simply doesn´t crash (and has no similar failures). It wasn´t my intention to enable all/many abandoned bots. Many weak bots never create an interesting match, often because they simply rush (5 pool, zealot, marine) in all of their games, and aren´t even that good at it.

Dan on :

I disagree(d) with that assessment that Hydra builds can deal with it. You can not answer four gate Goon with just moderate numbers of Hydralisks and expect to live. Bot Goons massacre Hydralisks in small numbers; you need to build up a number of Hydralisks to be able to push out against Dragoons, and when you finally do it needs to be decisive, all the way across the map without retreating. To even attempt doing this on two bases would require heavy Sunken turtling into five hatcheries + 30 drones, and this response is more all-in than the initial attack.

Lurkers are a tempting idea but a non-starter. You will never be able to leave your base as bot reaction times to Lurker burrow/unburrow are too fast and precise. If a Lurker manages to burrow it buys you space 32px at a time and you can't progress. When an Observer arrives, the game ends.

The strongest answer against one gate core strategies or 2-gate-into-4-gate is 2 hatch muta. Mutas force the Dragoons to pull back, allowing you to take a 3rd base (preferably another natural) and transition. The challenge is that it requires the Mutalisks to be effective at harassment; they need to get around the main army, and if cannons go up they need to be able to make smart decisions around them. Right now McRaveZ is the only bot I see starting to do this effectively.

Another lower-tech try is 5-6 hatchery pure speedling. This is what CherryPi ran with great efficacy against Locutus' FFE into five-gate Goon play in AIIDE 2018, but its success demands that the Zerglings engage efficiently, at least attempting to surround the Dragoon army rather than conga-lining in. A similar idea should work against four-gate Goon, using a few sunkens to provide an anchor against kiting while the Hatchery and Zergling counts rise.

Jay Scott on :

I agree about the hydras; I did not expect that build to work (though I’m glad to have the opening, which has other uses). I’m thinking that a speedling build is the way. I can work out the timings, but getting the tactics and micro right is not easy....

jtolmar on :

9pool speedlings looks like a plausible build against what I've seen of Stardust, even though 2gate is supposed to soft counter 9pool, since Stardusts's opening is a bit slower than the standard.

But I think this is over-emphasising builds over fundamentals. In the game Marco linked, between 5:00 and 7:00 Steamhammer harvests 1000 gas it doesn't use, and is too mineral-constrained to use all its larva. No strategy change is going to be as much of an improvement as just having an extra 1000 minerals worth of stuff in play.

MarcoDBAA on :

"they need to get around the main army, and if cannons go up they need to be able to make smart decisions around them. Right now McRaveZ is the only bot I see starting to do this effectively."

McRaveZ did this in the beginning of this game against PW:
http://www.openbw.com/replay-viewer/?rep=https://basilicum.bytekeeper.org/bots/McRaveZ/McRaveZ%20vs%20PurpleDestiny%20Andromeda%20CTR_B4F0D5EF.rep

Then feared the corsairs too much. Just spread the mutalisks out, and a few of them aren´t that troublesome. AI zerg should do better here than human zerg.

I also think that the sunkens at home were strong enough to not bring the mutas back into defense. Even if the goons try (and manage) to break through, they will lose enough of them to be in trouble vs the returning muta swarm, all while probes probably died at home before, which makes it more difficult to replenish forces.

Killerbot also has these muta skills, but all its mutas need to be in one squad to even attack the enemy. And it only plays one opening, which is not enough vs learning bots.

P.S: But I expect your PW bot to defeat Stardust in AIIDE, if the latter doesn´t get a strong update. www.openbw.com/replay-viewer/?rep=https://basilicum.bytekeeper.org/bots/Stardust/Stardust%20vs%20PurpleWave%20Icarus%20CTR_14A1D835.rep
A new bot always has the advantage to have tested vs everyone else before it was uploaded. This overestimates its true (potential) strength somewhat. Stardust still misses many skills. krasi0, even if not updated, is also defeating Stardust more often than not now. But sure, zerg is really in trouble at the moment.

Dan on :

I agree with your assessment of that McRaveZ-PW game. It's an example of McRaveZ's promise but also how much further it has to go. It navigates around the cannons and engages the natural mineral line (I think PurpleSwarm would fail at this, for example), but doesn't take all the opportunities it can. PW should have lost this game for multiple reasons, but getting value out of Mutalisks remains a challenge.

Killerbot's muta harassment is some of the best in bot-land -- second-best after McRaveZ I think -- but tends to eat too much damage while dishing too little of it out. As a result it fails to force honest play with respect to Mutalisk defense. For example, see http://www.openbw.com/replay-viewer/?rep=https://basilicum.bytekeeper.org/bots/Marian%20Devecka/Marian%20Devecka%20vs%20Ecgberht%20Icarus%20CTR_B3EC72DB.rep where Ecgberht gets away without building any turrets when it should need a whole bunch. Against one-base Protoss it doesn't healthily get into on-time Mutalisk production, as in http://www.openbw.com/replay-viewer/?rep=https://basilicum.bytekeeper.org/bots/Marian%20Devecka/Marian%20Devecka%20vs%20Randomhammer%20Fighting%20Spirit%20CTR_D4CAE8B6.rep

The PW-Stardust matchup is very close now, but I have no doubt Stardust will show up in stronger fighting shape than COG and will be very scary. I think Stardust is closer to winning the games it's losing than PurpleWave is in the game's Stardust is winning. And Stardust's win percentage against other opponents is very, very high; even if I win the heads-up matchup I will probably lose the round-robin. Will try my best for AIIDE but with only a couple of weeks to prepare (and an apartment move in between) I don't think I can get past Stardust.

Tully Elliston on :

Are you keeping the learning files from the ladder, or starting them afresh for the tourney?

Jay Scott on :

When both players are adapting to each other, it is not always good to start off a tournament with a completely trained learning file. That file tells what to do in the we-know-all-about-each-other state, which can be much different than the starting state, depending on the other bot’s behavior. Big startup transient = worse results.

MicroDK on :

It looks like Steamhammer is crashing on Fortress on the AIIDE ladder (Starcraft AI Ladder) after your latest update. ;)

Jay Scott on :

It was, but I fixed it already. Those tricky corner bases!

Add Comment

E-Mail addresses will not be displayed and will only be used for E-Mail notifications.

To prevent automated Bots from commentspamming, please enter the string you see in the image below in the appropriate input box. Your comment will only be submitted if the strings match. Please ensure that your browser supports and accepts cookies, or your comment cannot be verified correctly.
CAPTCHA

Form options

Submitted comments will be subject to moderation before being displayed.