Here’s my summary of BananaBrain’s learning files. BananaBrain records both its own strategy and the recognized enemy strategy for every game.
#1 stardust
| opening | games | wins | first | last |
|---|
| PvP_10/12gate | 5 | 0% | 9 | 121 |
| PvP_12nexus | 5 | 0% | 10 | 119 |
| PvP_2gatedt | 5 | 0% | 3 | 120 |
| PvP_2gatedtexpo | 16 | 19% | 0 | 125 |
| PvP_2gatereaver | 5 | 0% | 1 | 118 |
| PvP_3gaterobo | 5 | 0% | 13 | 123 |
| PvP_3gatespeedzeal | 5 | 0% | 5 | 116 |
| PvP_4gategoon | 24 | 25% | 12 | 126 |
| PvP_9/9gate | 5 | 0% | 6 | 122 |
| PvP_9/9proxygate | 38 | 29% | 8 | 156 |
| PvP_nzcore | 13 | 15% | 4 | 149 |
| PvP_zcore | 5 | 0% | 7 | 117 |
| PvP_zcorez | 9 | 11% | 11 | 144 |
| PvP_zzcore | 17 | 24% | 2 | 127 |
| 14 openings | 157 | 17% | | |
| enemy | games | wins |
|---|
| P_1gatecore | 6 | 50% |
| P_2gate | 26 | 19% |
| P_2gatefast | 13 | 31% |
| P_4gategoon | 107 | 14% |
| P_cannonturtle | 1 | 0% |
| P_unknown | 4 | 0% |
| 6 openings | 157 | 17% |
The most successful: Double proxy gates. Stardust plays the same every game, except for reactions to its opponent, so it’s interesting that BananaBrain diagnosed so many different openings. I suspect that they were all, or nearly all, 4 gate goon, and BananaBrain was not always able to scout long enough to see it. I think the variety is what you get when BananaBrain sees only part of the build.
#3 dragon
| opening | games | wins | first | last |
|---|
| PvT_10/12gate | 48 | 71% | 0 | 156 |
| PvT_1gatedtexpo | 1 | 0% | 16 | 16 |
| PvT_28nexus | 6 | 50% | 13 | 119 |
| PvT_2gaterngexpo | 4 | 50% | 10 | 91 |
| PvT_32nexus | 1 | 0% | 89 | 89 |
| PvT_9/9gate | 96 | 81% | 2 | 118 |
| PvT_9/9proxygate | 1 | 0% | 92 | 92 |
| 7 openings | 157 | 75% | | |
| enemy | games | wins |
|---|
| T_1fac | 59 | 66% |
| T_2fac | 1 | 100% |
| T_fastexpand | 1 | 100% |
| T_unknown | 96 | 79% |
| 4 openings | 157 | 75% |
The best builds were zealot builds. BananaBrain seems to be especially successful with early zealot pressure.
#4 steamhammer
| opening | games | wins | first | last |
|---|
| PvZ_10/12gate | 3 | 67% | 5 | 7 |
| PvZ_1basespeedzeal | 21 | 86% | 37 | 157 |
| PvZ_2basespeedzeal | 9 | 78% | 21 | 149 |
| PvZ_4gate2archon | 1 | 0% | 31 | 31 |
| PvZ_5gategoon | 2 | 50% | 29 | 30 |
| PvZ_9/9gate | 92 | 88% | 61 | 156 |
| PvZ_9/9proxygate | 1 | 0% | 57 | 57 |
| PvZ_bisu | 5 | 80% | 32 | 36 |
| PvZ_neobisu | 12 | 83% | 8 | 19 |
| PvZ_sairdt | 3 | 67% | 146 | 148 |
| PvZ_sairgoon | 1 | 0% | 20 | 20 |
| PvZ_sairreaver | 3 | 67% | 58 | 60 |
| PvZ_stove | 5 | 80% | 0 | 4 |
| 13 openings | 158 | 83% | | |
| enemy | games | wins |
|---|
| Z_10hatch | 32 | 84% |
| Z_12hatch | 57 | 75% |
| Z_12hatchmain | 1 | 100% |
| Z_12pool | 2 | 100% |
| Z_4/5pool | 1 | 100% |
| Z_9pool | 23 | 96% |
| Z_9poolspeed | 8 | 88% |
| Z_overpool | 19 | 84% |
| Z_unknown | 15 | 80% |
| 9 openings | 158 | 83% |
Again, zealot builds. Steamhammer tried a wide variety of counters, of which 12 hatch worked best. BananaBrain records only the earliest steps of zerg openings, so what BananaBrain calls Z_12hatch could have a range of followups.
#5 mcrave
| opening | games | wins | first | last |
|---|
| PvZ_10/12gate | 54 | 85% | 17 | 119 |
| PvZ_1basespeedzeal | 3 | 67% | 58 | 60 |
| PvZ_2basespeedzeal | 5 | 80% | 1 | 5 |
| PvZ_4gate2archon | 1 | 0% | 61 | 61 |
| PvZ_5gategoon | 1 | 0% | 66 | 66 |
| PvZ_9/9gate | 1 | 0% | 6 | 6 |
| PvZ_9/9proxygate | 8 | 75% | 24 | 100 |
| PvZ_bisu | 5 | 80% | 53 | 57 |
| PvZ_neobisu | 4 | 75% | 62 | 65 |
| PvZ_sairdt | 3 | 67% | 14 | 16 |
| PvZ_sairgoon | 12 | 83% | 7 | 105 |
| PvZ_sairreaver | 1 | 0% | 0 | 0 |
| PvZ_stove | 59 | 88% | 31 | 156 |
| 13 openings | 157 | 82% | | |
| enemy | games | wins |
|---|
| Z_12hatch | 84 | 81% |
| Z_12pool | 2 | 0% |
| Z_9pool | 23 | 78% |
| Z_overpool | 45 | 89% |
| Z_unknown | 3 | 100% |
| 5 openings | 157 | 82% |
Most things worked against McRave, but especially tech openings. The earliest steps of McRave’s openings are stereotyped, so BananaBrain recognized few choices.
#6 willyt
| opening | games | wins | first | last |
|---|
| PvT_10/12gate | 44 | 93% | 7 | 50 |
| PvT_12nexus | 6 | 83% | 0 | 5 |
| PvT_2gatedt | 1 | 0% | 6 | 6 |
| PvT_32nexus | 24 | 88% | 51 | 74 |
| PvT_9/9proxygate | 77 | 99% | 80 | 156 |
| PvT_dtdrop | 2 | 50% | 78 | 79 |
| PvT_stove | 3 | 67% | 75 | 77 |
| 7 openings | 157 | 93% | | |
| enemy | games | wins |
|---|
| T_1fac | 12 | 100% |
| T_2rax | 55 | 95% |
| T_fastexpand | 52 | 88% |
| T_unknown | 38 | 95% |
| 4 openings | 157 | 93% |
The proxy gates won 76 times out of 77. Ouch.
#7 microwave
| opening | games | wins | first | last |
|---|
| PvZ_10/12gate | 86 | 97% | 31 | 156 |
| PvZ_1basespeedzeal | 1 | 0% | 23 | 23 |
| PvZ_2basespeedzeal | 2 | 50% | 19 | 20 |
| PvZ_4gate2archon | 2 | 50% | 24 | 25 |
| PvZ_5gategoon | 30 | 83% | 39 | 79 |
| PvZ_9/9gate | 15 | 80% | 9 | 82 |
| PvZ_9/9proxygate | 2 | 50% | 63 | 64 |
| PvZ_bisu | 2 | 50% | 7 | 8 |
| PvZ_neobisu | 2 | 50% | 21 | 22 |
| PvZ_sairdt | 5 | 80% | 0 | 4 |
| PvZ_sairgoon | 5 | 80% | 26 | 30 |
| PvZ_sairreaver | 2 | 50% | 5 | 6 |
| PvZ_stove | 3 | 67% | 16 | 18 |
| 13 openings | 157 | 87% | | |
| enemy | games | wins |
|---|
| Z_10hatch | 8 | 100% |
| Z_12hatch | 31 | 97% |
| Z_12pool | 13 | 85% |
| Z_4/5pool | 13 | 100% |
| Z_9pool | 58 | 79% |
| Z_9poolspeed | 6 | 100% |
| Z_overpool | 20 | 75% |
| Z_unknown | 8 | 88% |
| 8 openings | 157 | 87% |
Zealots were best again, though dragoons were good too. I wonder why the economic 10/12 gates were more successful than the fast 9/9 gates? It suggests that Microwave may overdefend, fearing fast zealots, and not have a strong enough economy to hold off efficient zealots instead. Or the followup after the zealots; BananaBrain likes to expand quickly.
#8 daqin
| opening | games | wins | first | last |
|---|
| PvP_2gatedt | 10 | 80% | 0 | 37 |
| PvP_2gatedtexpo | 1 | 0% | 6 | 6 |
| PvP_2gatereaver | 142 | 92% | 7 | 156 |
| PvP_9/9gate | 3 | 67% | 31 | 33 |
| PvP_zcore | 1 | 0% | 26 | 26 |
| 5 openings | 157 | 90% | | |
| enemy | games | wins |
|---|
| P_1gatecore | 69 | 88% |
| P_4gategoon | 68 | 91% |
| P_ffe | 1 | 100% |
| P_unknown | 19 | 89% |
| 4 openings | 157 | 90% |
DaQin was apparently not ready for reavers. Otherwise it did not badly against a powerful opponent.
#9 freshmeat
| opening | games | wins | first | last |
|---|
| PvZ_4gate2archon | 8 | 88% | 26 | 33 |
| PvZ_9/9gate | 122 | 100% | 35 | 156 |
| PvZ_neobisu | 14 | 86% | 0 | 13 |
| PvZ_sairgoon | 1 | 0% | 34 | 34 |
| PvZ_stove | 12 | 83% | 14 | 25 |
| 5 openings | 157 | 96% | | |
| enemy | games | wins |
|---|
| Z_12hatch | 27 | 85% |
| Z_12hatchmain | 22 | 91% |
| Z_12pool | 1 | 100% |
| Z_4/5pool | 27 | 100% |
| Z_9pool | 11 | 100% |
| Z_overpool | 3 | 100% |
| Z_unknown | 66 | 100% |
| 7 openings | 157 | 96% |
#10 ualbertabot
| opening | games | wins | first | last |
|---|
| PvU_10/12gate | 4 | 75% | 0 | 3 |
| PvU_9/9gate | 1 | 0% | 4 | 4 |
| PvU_9/9proxygate | 5 | 80% | 10 | 14 |
| PvU_nzcore | 5 | 80% | 5 | 9 |
| PvU_zcore | 142 | 97% | 15 | 156 |
| 5 openings | 157 | 95% | | |
| enemy | games | wins |
|---|
| P_1gatecore | 19 | 100% |
| P_2gate | 1 | 100% |
| P_2gatefast | 25 | 84% |
| P_4gategoon | 3 | 100% |
| P_unknown | 6 | 100% |
| T_1fac | 1 | 100% |
| T_2fac | 22 | 100% |
| T_2rax | 16 | 94% |
| T_unknown | 11 | 100% |
| Z_12hatch | 26 | 100% |
| Z_4/5pool | 23 | 87% |
| Z_overpool | 3 | 100% |
| Z_unknown | 1 | 100% |
| 13 openings | 157 | 95% |
Comments
Dan on :
Major taxonomic question of PvP: is a build that gets 4 gateways and dragoons, but hits later and softer than a pure 4-gate goon due to inserting other stuff, still a 4-gate goon?
It's already quite difficult for a player to distinguish a 4-gate from a 3-gate which is an expand build, and which calls for a very different level of caution. Failing to see one of the dragoons of the first wave of 4 is enough to cause mis-identification. It's hard to reason about the absence of the dragoon based on confidence that you *would* have seen it if it existed.
Stardust likes to insert reactive/prophylactic things that render its 4-gate later than the traditional timing: a second gateway, a forge and cannon, etc. So even if it winds up getting 4 Gateways eventually, the build may function closer to a lower-pressure build like a 3-gate than a 4-gate with respect to how much you need to do on defense. Especially true if BB is opening proxy a lot, which if scouted would induce Stardust to add that second Gateway
Jay Scott on :
Jay Scott on :
It’s noticeable. I always thought that BananaBrain favored exploration over exploitation (which has more to do with the exploration parameter to UCB1 than the algorithm itself). The effect was unpredictability, making BananaBrain hard to exploit. These tables show a concentration on exploitation.