updates on Iron and Bereaver
Here are a couple of updates on top bots before I return to Overkill’s reinforcement learning.
Iron
For a while now, Iron has been playing a different strategy against protoss, going 2 barracks plus academy and attacking early with marines and medics rather than its traditional vultures. I’m a bit slow, but I finally realized why: It was having trouble against dragoons. Dragoons, adding observers in time to detect mines, are a strategically sound counter to Iron’s early vulture play, where tanks come late. Bereaver was handing Iron a string of losses with dragoon play. Time to switch it up!
Bereaver poked with 2 zealots but was unable to land a single hit against accurate micro.
Bereaver expanded early. I thought its force of 4 dragoons plus probes could have held off the marines, but it would have needed coordination and micro beyond the state of the art. Instead Bereaver lost its natural and was set back.
Iron tried to force the ramp while expanding, but didn’t have a big enough edge and was pushed back. Bereaver counterattacked and tried to get ahead with a double expansion, but tanks came out and Bereaver was never able to catch up. Bereaver strangely made a robo support bay early but did not build any reavers until much later, adding to its disadvantage—this game was played before the reupload below.
The dragoon opening is still sound against barracks play. Dragoons are versatile. But fighting the infantry calls for excellent micro with focus fire, and even Bereaver is lacking. Skynet is fairly skilled, but Iron’s new opening smashes it.
Bereaver
Bereaver was reuploaded yesterday. The new version’s first game was against Tyr by Simon Prins, and I immediately saw plays I hadn’t seen it make before. Bereaver went double robo.
Bereaver’s shuttle-reaver micro is not fluid like ZerGreenBot’s, but sharply goal-directed. Tyr vainly struggled to save itself, not realizing that it had to cover the reaver drop zones and come up with some kind of air defense. Every time Tyr unsieged, the reaver dropped, fired, was picked up instantly, the scarab skirted around the minerals—boom. By the time the second shuttle appeared with 1 zealot inside, it was already too late for terran.
The tanks at the top are approaching, too late to defend. As for reaver shots, sometimes they dud out. And sometimes they do this.
Reaver drop is crazy complicated. In time, I expect Bereaver to learn how to drop in the teeth of sieged tanks: Drop a zealot first. Tanks fire. Zealot says “ow!” Reaver drops and has time to fire before the tanks can reload. Terran defense has to be on the ball!
Comments
Jay Scott on :
Igor Dimitrijevic on :
Iron's current build against Protoss is a direct implementation of so called "Shallow 2" (I watched https://www.youtube.com/watch?v=iyfbhgcJ-Zo and liked the music).
Neither the build, nor the transition to Iron's default mid game, is adaptive for now. With good adaptive play during the early game though (add more barracks, expand later or sooner...), I expect the build to beat current Protoss strategies involving Dragoons and Zealots only.
There are several reasons I am currently testing bio builds against Protoss:
- First I can play with bio units.
- Variety: we lack bio Terran bots. Even bots that usually go bio against Terrans choose the mech way against Protoss.
- The matchup mech Terran vs Protoss seems well balanced for humans. But when it comes to bots... well, as a Protoss, you almost just need Dragoons. Add Zealots to easily counter sieged tanks. Both Dragoons and zealots are easy to micro, and the plan is straightforward for both: Zealots to Tanks, and Dragoons to anything. By contrast, both Tanks and Vultures require complex micro to be effective, and both require positionnal skills which are hard to implement, even if natural for a human player.
- In the matchup mech Terran vs Protoss, it seems Terran has to be defensive for a while. Even if Terrans are very good at this, I don't like it... And I think bots are more effective at attacking then defending. Going bio first turns this around, and that fits better with Iron's style of play.
- Also, going bio means lots of units available. As a general rule, I think bots are good at controling many units.
Jay Scott on :
LetaBot on :
krasi0 on :
Igor, keep up the good fight! :)