archive by month
Skip to content

AIIDE 2019 - what Locutus learned

Locutus’s game records are in almost the same format as Steamhammer’s and can be summarized by the same script. I expect it will also work for DaQin and AITP.

Where Steamhammer was set to keep 100 game records per opponent, Locutus was set to keep 200. Since there were 100 rounds in the tournament, game counts over 100 mean that pre-learned data is included in the table alongside the tournament data. If Locutus was not trained on a near-final version of the opponent, then the two could be significantly different.


#2 purplewave

openinggameswins
4GateGoon2854%
4GateGoonWithObs1662%
FakeDTRush1020%
ForgeExpand1963%
ZealotDrop12773%
5 openings20066%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Dark templar2010%50%3015%37%45%0%
Fast rush32%100%52%80%0%0%
Heavy rush63%17%126%67%0%0%
Not fast rush17186%69%15376%71%81%0%
timing#medianearlylate
gas steal attempt921:440:442:01
gas steal success15---
enemy scout1862:271:0916:11
enemy combat units1983:292:197:26
enemy air units556:504:5020:31
enemy cloaked units9311:075:1319:54


After seeing a few Locutus-PurpleWave games I got the impression that PurpleWave reacted adequately to Locutus’s trick strategy of cannoning the ramp and then dropping zealots. So I was surprised that Locutus considered it the best choice. But the overall win rate is high compared to the tournament results, so I suspect it is influenced by pre-learned data from games against a weaker version of PurpleWave.


#3 bananabrain

openinggameswins
4GateGoon1283%
ForgeExpand3784%
ZealotDrop15195%
3 openings20092%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Dark templar52%80%2814%100%40%0%
Fast rush63%100%126%83%0%0%
Heavy rush10%100%1910%89%0%0%
Not fast rush18894%92%14170%91%70%0%
timing#medianearlylate
gas steal attempt591:450:461:52
gas steal success6---
enemy scout1961:570:4610:09
enemy combat units2003:302:187:25
enemy air units2015:4113:0517:35
enemy cloaked units316:135:4616:11

#4 daqin

openinggameswins
4GateGoon1164%
FakeDTRush10%
ForgeExpand10%
ZealotDrop8787%
4 openings10083%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Dark templar99%89%2121%76%11%0%
Fast rush22%100%33%67%0%0%
Not fast rush8888%82%7676%86%74%0%
Unknown11%100%--0%0%
timing#medianearlylate
gas steal attempt261:450:451:49
gas steal success4---
enemy scout973:022:1518:05
enemy combat units1003:312:195:22
enemy air units217:4716:2319:10
enemy cloaked units917:186:029:17

#5 steamhammer

openinggameswins
4GateGoon7100%
9-9GateDefensive5100%
CannonFirst4GateGoon11100%
ForgeExpand4Gate2Archon1173%
ForgeExpand5GateGoon15595%
ForgeExpandSpeedlots1100%
PlasmaCorsairsCarriers9100%
ProxyHeavyZealotRush210%
8 openings20094%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Fast rush147%100%2613%96%57%0%
Heavy rush2613%81%5930%92%38%0%
Hydra bust42%100%2312%87%50%0%
Not fast rush15678%96%9246%98%53%0%
timing#medianearlylate
gas steal attempt0---
gas steal success0---
enemy scout1882:210:5118:45
enemy combat units1993:102:027:11
enemy air units125:405:016:14
enemy cloaked units710:025:1519:39


The numbers in the “recognized” columns of the plan table show how widely Steamhammer cast its net for a solution to Locutus.

Locutus never tried to steal the gas of a zerg. Objectively, that makes sense. In the context of bot play, I’m not so sure; many bots of all races mess up their builds in the face of a gas steal.


#6 zzzkbot

openinggameswins
9-9GateDefensive4100%
CannonAtChokeFirst4GateGoon1354%
CannonFirst4GateGoon17899%
PlasmaCorsairsCarriers1100%
PlasmaProxy2Gate4100%
5 openings20096%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Fast rush11557%97%10754%94%73%0%
Heavy rush7839%96%6633%100%54%0%
Hydra bust10%100%21%100%100%0%
Not fast rush63%100%2512%96%17%0%
timing#medianearlylate
gas steal attempt0---
gas steal success0---
enemy scout2002:250:515:58
enemy combat units1962:282:037:59
enemy air units527:535:2613:43
enemy cloaked units0---

#7 microwave

openinggameswins
9-9GateDefensive2100%
ForgeExpand4Gate2Archon367%
ForgeExpand5GateGoon14699%
ForgeExpandSpeedlots4480%
PlasmaCorsairsCarriers2100%
PlasmaProxy2Gate3100%
6 openings20094%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Fast rush5025%82%4120%98%32%0%
Heavy rush3417%100%5025%90%47%0%
Hydra bust--178%100%0%0%
Not fast rush11557%98%9146%95%57%0%
Proxy10%100%10%100%0%0%
timing#medianearlylate
gas steal attempt0---
gas steal success0---
enemy scout1952:301:0721:14
enemy combat units1983:031:477:59
enemy air units6911:195:4924:25
enemy cloaked units326:375:2113:49

#8 iron

openinggameswins
CautiousDTDrop20098%
1 openings20098%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Not fast rush136%100%6130%100%46%0%
Wall-in18794%98%13970%97%71%0%
timing#medianearlylate
gas steal attempt350:460:450:48
gas steal success11---
enemy scout1902:451:4210:46
enemy combat units2004:072:346:39
enemy air units1178:186:5513:39
enemy cloaked units1178:186:5513:39


Locutus declared an enemy-specific strategy against Iron. I’m not sure why it also had pre-learned data.


#9 xiaoyi

openinggameswins
10-15GateGoon10%
10Gate25NexusFE250%
DTDrop10%
ForgeExpand10%
Proxy2ZealotsIntoGoons3093%
ProxyDTRush16595%
6 openings20093%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Not fast rush200100%93%200100%93%100%0%
timing#medianearlylate
gas steal attempt681:171:121:50
gas steal success12---
enemy scout1943:012:1115:29
enemy combat units2004:202:296:57
enemy air units1312:597:5415:10
enemy cloaked units48:007:548:18


Proxy DT rush. That tends to confirm my picture of XiaoYi as vulnerable to tricks.


#10 mcrave

openinggameswins
4GateGoon580%
4GateGoonWithObs3100%
FakeDTRush10%
ForgeExpand250%
ZealotDrop18994%
5 openings20093%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Dark templar32%100%168%94%67%0%
Fast rush32%100%189%100%0%0%
Heavy rush21%50%52%100%0%0%
Not fast rush19296%93%16180%92%82%0%
timing#medianearlylate
gas steal attempt991:460:451:57
gas steal success2---
enemy scout1932:091:2114:38
enemy combat units2003:352:217:26
enemy air units2411:447:1920:31
enemy cloaked units6011:035:1514:25

#11 ualbertabot

openinggameswins
CannonFirst4GateGoon18899%
PlasmaProxy2Gate10100%
Proxy9-9Gate20%
3 openings20098%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Dark templar10%100%116%100%0%0%
Fast rush2312%100%2914%93%22%0%
Heavy rush4824%96%6332%100%29%0%
Not fast rush12864%99%9748%99%51%0%
timing#medianearlylate
gas steal attempt952:001:572:03
gas steal success0---
enemy scout1272:111:185:45
enemy combat units1353:222:016:54
enemy air units76:456:416:53
enemy cloaked units114:344:305:13


Locutus configured an enemy-specific strategy against UAlbertaBot. Openings other than CannonFirst4GateGoon are from pre-learned data, which was ignored in making the opening decision.


#12 aitp

openinggameswins
DTDrop66100%
ForgeExpand3397%
Turtle1100%
3 openings10099%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Fast rush22%100%77%100%0%0%
Heavy rush11%100%11%100%0%0%
Not fast rush7777%99%6262%100%62%0%
Unknown11%100%--0%0%
Wall-in1919%100%3030%97%42%0%
timing#medianearlylate
gas steal attempt430:460:451:27
gas steal success15---
enemy scout263:192:416:02
enemy combat units1003:482:017:49
enemy air units0---
enemy cloaked units0---

#13 bunkerboxer

openinggameswins
10Gate25NexusFE2195%
CannonFirst4GateGoon88100%
ForgeExpand79100%
PlasmaProxy2Gate10100%
Proxy9-9Gate20%
5 openings20098%
planpredictedrecognizedaccuracy
countgameswinscountgameswinsgood?
Fast rush178%100%2010%90%24%0%
Heavy rush3618%94%3518%100%31%0%
Not fast rush14774%99%14472%99%83%1%
Unknown--10%100%0%0%
timing#medianearlylate
gas steal attempt871:290:452:03
gas steal success16---
enemy scout1042:071:273:18
enemy combat units1132:262:018:14
enemy air units0---
enemy cloaked units0---

overall

totalPvTPvPPvZPvR
openinggameswinsgameswinsgameswinsgameswinsgameswins
10-15GateGoon10% 10%
10Gate25NexusFE2391% 2391%
4GateGoon6368% 5664% 7100%
4GateGoonWithObs1968% 1968%
9-9GateDefensive11100% 11100%
CannonAtChokeFirst4GateGoon1354% 1354%
CannonFirst4GateGoon465100% 18999% 276100%
CautiousDTDrop20098% 20098%
DTDrop6799% 6799%
FakeDTRush1217% 1217%
ForgeExpand17290% 11398% 5975%
ForgeExpand4Gate2Archon1471% 1471%
ForgeExpand5GateGoon30197% 30197%
ForgeExpandSpeedlots4580% 4580%
PlasmaCorsairsCarriers12100% 12100%
PlasmaProxy2Gate27100% 7100% 20100%
Proxy2ZealotsIntoGoons3093% 3093%
Proxy9-9Gate40% 40%
ProxyDTRush16595% 16595%
ProxyHeavyZealotRush210% 10%
Turtle1100% 1100%
ZealotDrop55488% 55488%
total220092%60097%70084%60095%30098%
openings played2285103

Trackbacks

No Trackbacks

Comments

Dan on :

The drop opener dynamic vs. PurpleWave is trickier than it looks.

On first glance, yes, PurpleWave sees the cannons and reacts appropriately. There's a lot of nonsense you can do (and bots do do) behind cannons (Carrier rush, DT rush, fast expansion, 4-Gate Goon, 4-gate proxy mass slowlot), and PurpleWave's reaction is sound against them all (expansion into Observer + Reaver).

PurpleWave also generally comes out of the drop itself in good shape. It could do a better job of saving probes but it still tends to wind up ahead economically.

One little advantage is adding noise to PurpleWave's learning data. Because that cannon reaction can come out of any opener, and PurpleWave's learning doesn't understand that a deviation happened, you make it harder for PurpleWave to learn what builds work against your bot as a whole. I think this effect is fairly minor.

A nastier bit about the drop build is that it forced PurpleWave into a Shuttle+Reaver composition, which was new in AIIDE and still shaky. So shaky that Locutus could frequently overcome its build order disadvantage by virtue of PurpleWave throwing its Reavers.

The other nasty bit -- and the more significant one, I think -- is that it forces PurpleWave to throw out its well-oiled opening book for PvP. PurpleWave thrives on gaining early-game advantages by detecting what you're doing and reacting correctly. This goes down to the resolution of "do I even bother making Zealots before my first Dragoon?" -- it really tries to get every edge and cut every corner it can. But opening cannons means it's settling for the moderate-but-not-lethal advantage of having a faster expansion and better (in theory, but frankly worse than pure Dragoon in execution) tech. And the damage dealt by the drop comes close to evening that advantage back out.

So mainly it's the combination of guaranteed early game survival vs. PurpleWave + forcing out PurpleWave's mediocre Shuttles that let the drop opening make for a fair fight against PurpleWave.

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.