AIIDE 2019 - what DaQin learned

DaQin is derived from Locutus and also keeps 200 game records. But DaQin did not have pre-learned data. No games were left uncompleted; there are 100 against each opponent.

DaQin plays fewer builds than the other bots I’ve looked at so far.

#1 locutus

opening	games	wins
3GateDT	100	17%
1 openings	100	17%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
DarkTemplar rush	89	89%	16%	96	96%	17%	97%	2%
Proxy	6	6%	17%	2	2%	0%	0%	0%
Unknown	5	5%	40%	2	2%	50%	0%	0%

timing	#	median	early	late
gas steal attempt	47	1:43	1:39	2:06
gas steal success	0	-	-	-
enemy scout	99	6:07	1:21	9:07
enemy combat units	100	4:34	2:22	6:47
enemy air units	96	6:30	4:02	18:41
enemy cloaked units	0	-	-	-

DaQin had an enemy-specific strategy configured for Locutus, so it didn’t try anything else. Locutus is the only opponent that DaQin tried to prepare for, as far as I can see.

DaQin incorrectly recognized dark templar rush as Locutus’s strategy in most games, then correctly recorded that no cloaked units were seen during the game. See yesterday for Locutus’s play against DaQin, which did not include any DT build. I assume that the dark templar recognition is deliberately over-cautious, because DTs are dangerous. Locutus does have a fake dark templar build, where it adds a citadel of Adun to fool opponents into expecting dark templar (it works against most UAlbertaBot-derived bots).

#2 purplewave

opening	games	wins
2GateDT	23	22%
3GateDT	3	0%
4GateGoon	74	14%
3 openings	100	15%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
DarkTemplar rush	32	32%	16%	35	35%	23%	69%	0%
Fast rush	66	66%	14%	64	64%	11%	80%	0%
Proxy	1	1%	100%	1	1%	0%	0%	0%
Unknown	1	1%	0%		-	-	0%	0%

timing	#	median	early	late
gas steal attempt	29	0:46	0:46	0:50
gas steal success	8	-	-	-
enemy scout	99	2:17	1:18	4:41
enemy combat units	99	2:47	2:21	5:13
enemy air units	41	8:42	4:05	18:10
enemy cloaked units	85	6:07	5:06	15:41

Against PurpleWave, in contrast, DaQin less often foresaw dark templar, but apparently often faced them. (Arbiters can’t get out that fast.)

#3 bananabrain

opening	games	wins
2GateDT	4	25%
3GateDT	68	56%
4GateGoon	28	36%
3 openings	100	49%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
DarkTemplar rush	47	47%	53%	55	55%	62%	51%	0%
Fast rush	48	48%	44%	39	39%	33%	35%	0%
Heavy rush	1	1%	0%	2	2%	50%	0%	0%
Not fast rush	1	1%	100%	2	2%	0%	0%	0%
Proxy	1	1%	100%	2	2%	50%	0%	0%
Unknown	2	2%	50%		-	-	0%	0%

timing	#	median	early	late
gas steal attempt	43	1:42	0:46	1:48
gas steal success	9	-	-	-
enemy scout	100	1:59	1:21	3:09
enemy combat units	100	2:57	2:19	5:43
enemy air units	67	8:14	3:58	12:42
enemy cloaked units	28	5:47	4:57	19:38

#5 steamhammer

opening	games	wins
ForgeExpand5GateGoon	100	94%
1 openings	100	94%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Fast rush		-	-	1	1%	100%	0%	0%
Heavy rush	29	29%	97%	18	18%	100%	14%	3%
Hydra bust	1	1%	100%	2	2%	100%	0%	0%
Not fast rush	64	64%	92%	72	72%	93%	69%	8%
Proxy		-	-	1	1%	100%	0%	0%
Unknown	6	6%	100%	6	6%	83%	0%	0%

timing	#	median	early	late
gas steal attempt	0	-	-	-
gas steal success	0	-	-	-
enemy scout	97	2:25	0:51	6:03
enemy combat units	100	3:17	1:57	7:03
enemy air units	18	9:23	5:30	16:18
enemy cloaked units	16	5:51	4:57	13:43

#6 zzzkbot

opening	games	wins
ForgeExpand5GateGoon	97	10%
ForgeExpandSpeedlots	3	0%
2 openings	100	10%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Fast rush	3	3%	33%	5	5%	100%	0%	33%
Heavy rush	90	90%	3%	93	93%	4%	100%	0%
Not fast rush		-	-	1	1%	100%	0%	0%
Unknown	7	7%	86%	1	1%	0%	0%	0%

timing	#	median	early	late
gas steal attempt	0	-	-	-
gas steal success	0	-	-	-
enemy scout	97	2:57	0:59	7:30
enemy combat units	100	2:39	1:47	4:31
enemy air units	7	7:58	7:46	8:25
enemy cloaked units	0	-	-	-

How did ZZZKBot upset DaQin? These numbers suggest zergling bust (it could be hydras, but DaQin does have a hydra bust recognizer which did not fire): Mostly “heavy rush,” few mutalisks, no lurkers. Steamhammer also settled on zergling bust as the best bet, but was much less successful. Microwave tried its zergling bust build versus DaQin without success. Maybe ZZZKBot’s extreme aggression is the key.

#7 microwave

opening	games	wins
ForgeExpand5GateGoon	84	85%
ForgeExpandSpeedlots	16	75%
2 openings	100	83%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Fast rush	15	15%	93%	15	15%	100%	33%	0%
Heavy rush	32	32%	81%	20	20%	85%	16%	9%
Not fast rush	50	50%	80%	59	59%	76%	66%	4%
Proxy		-	-	1	1%	100%	0%	0%
Unknown	3	3%	100%	5	5%	100%	0%	0%

timing	#	median	early	late
gas steal attempt	0	-	-	-
gas steal success	0	-	-	-
enemy scout	97	2:33	1:10	6:10
enemy combat units	90	3:29	1:50	6:37
enemy air units	41	10:37	5:15	14:07
enemy cloaked units	5	6:31	6:23	10:23

#8 iron

opening	games	wins
12NexusCarriers	92	96%
4GateGoon	8	50%
2 openings	100	92%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Factory	55	55%	96%	95	95%	94%	95%	2%
Proxy	8	8%	50%	3	3%	33%	0%	12%
Unknown	37	37%	95%	2	2%	100%	0%	0%

timing	#	median	early	late
gas steal attempt	92	2:19	2:15	2:35
gas steal success	0	-	-	-
enemy scout	87	2:58	1:41	12:09
enemy combat units	100	4:18	2:50	5:49
enemy air units	36	8:23	6:29	15:43
enemy cloaked units	30	8:25	7:54	15:43

12NexusCarriers seems to be the default build versus terran. Apparently terrans, even Iron, were not able to punish the fast expand. Well, they’re not supposed to be able to without risk, that’s the point of cutting probes for nexus on 12, but it does require good play from protoss to ensure.

#9 xiaoyi

opening	games	wins
12NexusCarriers	93	84%
3GateDT	1	0%
4GateGoon	1	0%
DTDrop	5	80%
4 openings	100	82%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Factory	60	60%	83%	47	47%	94%	48%	42%
Not fast rush	29	29%	76%	10	10%	80%	14%	48%
Proxy	1	1%	0%	2	2%	100%	0%	0%
Safe expand	4	4%	100%	1	1%	0%	0%	0%
Unknown	6	6%	100%	40	40%	70%	0%	17%

timing	#	median	early	late
gas steal attempt	99	2:19	0:46	2:25
gas steal success	3	-	-	-
enemy scout	93	2:23	2:10	19:03
enemy combat units	100	3:24	2:33	7:06
enemy air units	80	8:23	7:09	17:30
enemy cloaked units	11	8:15	7:57	8:27

XiaoYi usually got air tech pretty fast, that’s unusual and interesting. I’m guessing it scouted the carriers coming and prepared wraiths.

#10 mcrave

opening	games	wins
2GateDT	1	0%
3GateDT	62	52%
4GateGoon	37	24%
3 openings	100	41%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
DarkTemplar rush	19	19%	53%	16	16%	62%	63%	0%
Fast rush	79	79%	39%	83	83%	36%	96%	0%
Naked expand		-	-	1	1%	100%	0%	0%
Unknown	2	2%	0%		-	-	0%	0%

timing	#	median	early	late
gas steal attempt	13	1:41	0:46	1:46
gas steal success	2	-	-	-
enemy scout	100	2:22	1:25	6:11
enemy combat units	100	3:03	2:21	5:29
enemy air units	21	6:11	3:38	15:57
enemy cloaked units	76	6:23	5:17	8:33

McRave upset DaQin. Dark templar in 3 out of 4 games, and they came out pretty early. PurpleWave showed a similar pattern, but it wasn’t as salient because it wasn’t an upset. The dark templar rush recognizer did not seem to be fully effective, possibly because it was overridden by the fast rush recognizer. DaQin’s best counter was DT-back-atcha.

#11 ualbertabot

opening	games	wins
12NexusCarriers	2	50%
3GateDT	25	88%
4GateGoon	4	50%
DTDrop	2	50%
ForgeExpand5GateGoon	67	78%
5 openings	100	78%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
DarkTemplar rush	12	12%	75%	11	11%	91%	17%	8%
Factory	3	3%	67%	11	11%	100%	0%	0%
Fast rush	67	67%	78%	47	47%	57%	51%	7%
Heavy rush	1	1%	100%	5	5%	100%	0%	0%
Hydra bust		-	-	1	1%	100%	0%	0%
Not fast rush	13	13%	92%	15	15%	100%	8%	15%
Proxy	1	1%	0%	2	2%	50%	0%	0%
Unknown	3	3%	67%	8	8%	100%	0%	0%

timing	#	median	early	late
gas steal attempt	24	1:43	0:46	2:17
gas steal success	2	-	-	-
enemy scout	87	1:47	1:14	9:30
enemy combat units	98	3:01	1:38	6:58
enemy air units	9	7:37	6:07	15:47
enemy cloaked units	4	5:09	4:33	5:19

DaQin had some trouble adapting to random UAlbertaBot. This is a point where preparation for the opponent would have been valuable: Make a build that UAlbertaBot can’t beat and ensure that it is played. It can be a general-purpose build; PurpleWave included a cannon turtle build that is safe against all sorts of rushes.

#12 aitp

opening	games	wins
12NexusCarriers	100	100%
1 openings	100	100%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Factory	79	79%	100%	5	5%	100%	4%	96%
Unknown	21	21%	100%	95	95%	100%	0%	90%

timing	#	median	early	late
gas steal attempt	100	2:19	2:16	2:25
gas steal success	0	-	-	-
enemy scout	11	7:53	2:38	11:45
enemy combat units	100	5:55	2:43	7:29
enemy air units	67	10:07	8:50	14:01
enemy cloaked units	0	-	-	-

#13 bunkerboxer

opening	games	wins
12NexusCarriers	95	98%
4GateGoon	5	100%
2 openings	100	98%

plan	predicted			recognized			accuracy
plan	count	games	wins	count	games	wins	good	?
Fast rush		-	-	1	1%	100%	0%	0%
Not fast rush	78	78%	97%	35	35%	100%	35%	60%
Proxy	5	5%	100%	5	5%	100%	0%	0%
Unknown	17	17%	100%	59	59%	97%	0%	71%

timing	#	median	early	late
gas steal attempt	93	2:20	2:15	7:19
gas steal success	28	-	-	-
enemy scout	62	2:07	1:47	7:18
enemy combat units	59	2:59	2:09	7:51
enemy air units	0	-	-	-
enemy cloaked units	0	-	-	-

Beating BunkerBoxeR with a build of fast expansion into carriers is... not the intuitive choice. But I guess it worked.

overall

	total		PvT		PvP		PvZ		PvR
opening	games	wins	games	wins	games	wins	games	wins	games	wins
12NexusCarriers	382	94%	380	94%					2	50%
2GateDT	28	21%			28	21%
3GateDT	259	42%	1	0%	233	37%			25	88%
4GateGoon	157	25%	14	64%	139	21%			4	50%
DTDrop	7	71%	5	80%					2	50%
ForgeExpand5GateGoon	348	65%					281	62%	67	78%
ForgeExpandSpeedlots	19	63%					19	63%
total	1200	63%	400	93%	400	30%	300	62%	100	78%
openings played	7		4		3		2		5

Trackbacks

No Trackbacks

Comments

Dan on Wednesday, October 30. 2019:

DaQin was losing to the 4pool build, even while opening with cannons.

It's shockingly hard to reliably beat 4-pool with cannons. There are two approaches:

1. Open blindly with 8 Pylon 9 Forge 9 Cannon 9 Cannon. This can get cannons up in time, but you can not afford to scout (you need everyone mining). Thus it's terrible against macro builds -- it's four workers behind with no scouting information. If they open 12 Hatch, you are very behind.

2. Open with 8 Pylon, 10/11 Forge, 12/13 Cannon 12/13 Cannon. You scout with your Pylon builder. If you see the 4-pool coming, you pull your workers to defend the cannon. Failure to pull the Probes results in death. Last-scouting on a 4-player map results in death. Letting the Zerglings squeeze past the Probes into the cannons results in death. Pulling the Probes before having 300 minerals for your two cannons results in death. Pulling them three seconds later than that results in death. The timing is incredibly tight.

DaQin opens 8 Pylon 9 Forge 9 Cannon 9 Cannon, sometimes in-main. But it wasn't fast enough. In one game I watched, it was 16 seconds behind the fastest possible cannons, which was fatal: 6 seconds due to scouting (delaying the 150 Forge minerals), and 10 seconds due to delayed cannon placement (partly due to lacking the 300 Cannon minerals, in turn due to scouting). Even though that was in-main, it was too late and the cannons weren't complete, resulting in death. Scouting was the major error there, but the building placements could also have been a couple of seconds faster.

MarcoDBAA on Wednesday, October 30. 2019:

I think, that it makes sense to always have extreme openings vs bots available. No-look turtling, super greedy, fast variable rushes...

Very possible, that it might be exactly what is needed.
Enough rush bots, that don´t have (many) other options (not as common anymore), or rush too often (objectively). There are even very strong bots like SAIDA, that had massive problems with invisible units. And I remember MadMix upsetting stronger (especially in micro) bots by going full economy and also recuperating well (even with losing the main). Jay also mentioned SH winning a few games vs Locutus with a macro opening.

Dan on Wednesday, October 30. 2019:

The downside is the cost of exploring diverse openings. In win% tournaments, the most costly situation is to have no builds that win, but the second-most-costly is to have builds that *sometimes* win.

MarcoDBAA on Thursday, October 31. 2019:

Sure, but if a bot can recognize an enemy plan, it may try a conservative approach at first, that still allows the bot to react to something else.

If this fails, even though the opponent still rushes or still turtles to get carriers (2 examples), that extreme approach might be the best bet already.

Add Comment

Name*

Homepage

Comment*

In reply to

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