OR  Zero-K Name:    Password:   

1v1 Matchup data for 9th to 17th of May

59 posts, 1305 views
Post comment
Filter:    Player:  
Page of 3 (59 records)

17 days ago
amphib is legit, look at some recent randy games hes been trying to make it a thing and it works

archer perma stun locks kodachis and does decently enough vs blitz also then u have grizz/buoy which destroys cyclops big time

he beat me on gecko isle quite decisively with amphib, which is a map where I beat him quite decisively beforehand with same lab (tanks)

definitely no issue with amphib fac
+0 / -0

17 days ago
I watched http://zero-k.info/Battles/Detail/547770 and while it did feel Archer was good against Kodachi, it seemed to fare poorly against Blitz. Grizz/Buoy being strong against Cyclops is known (I'd say that's what gave Randy the win there.
+0 / -0

16 days ago
the only good use for buoy is vs reapers anyway,if they attack lol
+0 / -0

13 days ago
(edited 13 days ago)

Updated data from 10th of May to 6th of June is here.

Some things i've tested so far, using elo cutoff = 2200 for both players:

Is ZK just a game of Tank-K at high level? (my expectation: "winrate slightly over 50%")

Tank games: 134
Tank wins: 75
Win percentage: 0.5597014925373134

Seems fine, Tan-K stronk! But what if we remove mirror? (all further data is with mirrors eliminated)

factorytank games: 109
factorytank wins: 50
Win rate: 0.45871559633027525


Okay, so what's pwning tank?

vs factoryveh: 50 battles, 22 wins, ratio 0.44
vs factoryshield: 14 battles, 7 wins, ratio 0.5
vs factorycloak: 15 battles, 8 wins, ratio 0.5333333333333333
vs factoryamph: 4 battles, 1 wins, ratio 0.25
vs factoryjump: 9 battles, 3 wins, ratio 0.3333333333333333
vs factoryspider: 3 battles, 0 wins, ratio 0
vs factoryhover: 8 battles, 3 wins, ratio 0.375
vs factorygunship: 5 battles, 5 wins, ratio 1
vs undefined: 1 battles, 1 wins, ratio 1

Cloaky and Shield are surprisingly even, gunships predictably lose. Hover, Rover and Amph are favored against Tank.

Okay, maybe Tan-K doesn't survive in the chilly heights over 2200 elo.
So what about the next contender, Amph-K with the horrible Archers and Grizzlies?

factoryamph games: 39
factoryamph wins: 22
Win rate: 0.5641025641025641
vs factoryshield: 6 battles, 1 wins, ratio 0.16666666666666666
vs factorytank: 4 battles, 3 wins, ratio 0.75
vs factoryhover: 3 battles, 2 wins, ratio 0.6666666666666666
vs factorygunship: 7 battles, 6 wins, ratio 0.8571428571428571
vs factorycloak: 9 battles, 6 wins, ratio 0.6666666666666666
vs factoryjump: 3 battles, 2 wins, ratio 0.6666666666666666
vs factoryveh: 6 battles, 2 wins, ratio 0.3333333333333333
vs factoryspider: 1 battles, 0 wins, ratio 0

This is looking like a much nastier threat to the world balance, hmm. Amphs are still quite unpopular though, and in this data they lose to rovers and shields. Speaking of rovers, how do these fare?

factoryveh games: 100
factoryveh wins: 59
Win rate: 0.59
vs factorytank: 50 battles, 28 wins, ratio 0.56
vs factorycloak: 12 battles, 7 wins, ratio 0.5833333333333334
vs factoryshield: 9 battles, 4 wins, ratio 0.4444444444444444
vs factoryhover: 8 battles, 5 wins, ratio 0.625
vs factoryjump: 11 battles, 7 wins, ratio 0.6363636363636364
vs factoryspider: 3 battles, 3 wins, ratio 1
vs factoryamph: 6 battles, 4 wins, ratio 0.6666666666666666
vs factoryplane: 1 battles, 1 wins, ratio 1

.... it has been Rover-K all along???
+2 / -0

12 days ago
(edited 12 days ago)

I really don't think we can draw any conclusions from games that are not 2500+(below this is E-stall-K) and <100 elo difference. Things get sooo lopsided around ~200 elo difference that balance doesn't matter for shit when the higher elo player just wins due to being so much better.

This data would have much more validity if it adjusted for player WHR and if that caused deviation from expected winrates.

Plus, with such a small sample size for each matchup, you can't honestly claim anything. Yes the winratio might be .64, but it is wrong to say that statistically and rather say its .64 +- .333 or whatever, giving a probability that the true value is in between .33 and .98 or whatever, making these numbers effectively useless.
+3 / -0

12 days ago
They are still more useful than the bare claim that tanks are op.
+3 / -0

12 days ago
Not sure if my brain is farting, but I don't see how removing mirror matches from the data can reduce win-rate (tank-k data set).

The population being removed is by definition 50% winrate no? So removing it from the 55% winrate data should increase overall winrate, not reduce it?
+0 / -0

12 days ago
Depends whether tank mirrors are counted as 1 or 2 tank games. I would expect 2.
+0 / -0

12 days ago
Each Tank v Tank game counted as 1 Tank game, therefore ensuring a 100% win rate and distorting figures. Not sure why they left such a bad figure in but w/e
+2 / -0

12 days ago
Ah, thanks. That clears it up.
+0 / -0

12 days ago
(edited 12 days ago)

As others have pointed out, EErankAdminAnarchid has processed the data incorrectly. We can use what they have written to correct the error.

There are 134 games in which tanks were picked and 75 of those games were won by a tank picker. In games where only one side picked tanks there were 50 wins.
  • Therefore thare 25 games in which both sides picked tanks.
  • Therefore tanks were picked 159 times overall.
  • Therefore the correct winrate for tanks is 75/159 = 0.472.

I'll make some tables from the new raw data shortly.
+2 / -0

12 days ago
(edited 12 days ago)

Here are more games. This set is from the 10th of May to the 5th of June. I have removed games in which one side failed to plop a factory.

All Games

Battles: 5196
Factory Winrate (excluding mirror) Pick Count Mirror Matches
Cloaky 48.41% (47.81%) 2448 332
Shield 50.15% (50.17%) 1001 65
Rover 50.14% (50.17%) 1436 134
Hover 49.20% (49.13%) 561 23
Spider 51.42% (51.71%) 1159 97
Jump 51.53% (51.71%) 885 48
Tank 53.87% (54.93%) 1743 187
Amph 52.96% (53.33%) 423 24
Plane 27.21% (26.57%) 147 2
Gunship 44.10% (43.58%) 585 24
Ship 0.00% (0.00%) 4 0


1500 Minimum WHR

Battles: 3225
Factory Winrate (excluding mirror) Pick Count Mirror Matches
Cloaky 49.49% (49.32%) 1364 169
Shield 50.89% (51.05%) 619 48
Rover 48.72% (48.34%) 934 106
Hover 50.29% (50.32%) 344 17
Spider 49.72% (49.65%) 716 74
Jump 49.22% (49.10%) 644 44
Tank 52.51% (53.33%) 1215 149
Amph 53.82% (54.39%) 327 21
Plane 42.22% (41.86%) 45 1
Gunship 41.91% (41.56%) 241 5
Ship 0.00% (0.00%) 1 0


2000 Minimum WHR

Battles: 875
Factory Winrate (excluding mirror) Pick Count Mirror Matches
Cloaky 46.56% (44.96%) 378 60
Shield 57.79% (59.52%) 154 14
Rover 53.57% (54.85%) 280 37
Hover 53.93% (54.55%) 89 6
Spider 47.71% (47.15%) 153 15
Jump 50.70% (50.78%) 142 7
Tank 48.95% (48.48%) 382 59
Amph 47.22% (46.59%) 108 10
Plane 16.67% (16.67%) 6 0
Gunship 48.28% (48.21%) 58 1

+3 / -0

12 days ago
Tank is dominating at lower levels, if you call 53.87% dominating, but it is also quite popular. Surprisingly, Tank and Amph are unfavoured at higher levels. A deeper look is required to see if there are systematic biases in the data. Perhaps people 'tryhard' when facing better players and so pick tanks.

I think it is safe to say that Tank is strong against Cloaky. As the two most common picks they see a lot of play and the matchup table says that Tank wins around 60% of the time. A fairly simple change would be to make Blitz 2-shot Glaive and, given the data, I think it is worth doing fairly soon.
+1 / -0

12 days ago
Irrespective of this data, Blitz OP
+0 / -0

12 days ago
I think nerfing blitz range would be healthier than a 50% nerf against glaive because range reduces the number of blitzes capable of firing simultaneously, slightly lowering its balling power. Also it lets riots do a bit more damage.

Alternatively, longer reload time would mean less glaives die per second also without requiring huge jumps like 2x shots needed.

OTOH, Kodachi reks glaives anyway.
+4 / -0

12 days ago
With a win rate of 48.95% in 382 matches, there's still a chance of 34% that tank wins a majority of games.
+3 / -0

12 days ago
(edited 11 days ago)

I was actually just searching posts to see how often you guys post game matchup data. So excited to see that you just added so much more. I've spent a lot of time since I first saw this - trying to analyze the statistical implications to try to figure out if it's relevant to balance or not. Its actually an really interesting question, and just the kind of thing I like mulling over.

The problem comes down largely to selection bias. If there was extra "mystery factory" button ingame to place your first factory as random (in the name of science of course) that would be a different story. (Now that I think about it, even then there would be plenty of selection bias because which map you felt like going random on, and how wasted you are when you choose random, and maybe tanks are easier to play wasted).

But even then, if the choice of "choosing random" was forced and randomized, even then you'd still have plenty of factors like gameplay-meta and preconceptions-of-the-day affecting gameplay, although I suppose that just needs to be accepted as unavoidable.

Before I go any further, I have to say that it's extremely impressive how close to 50% these are. 53% considered dominating? I think you can call that a success and give yourself a pat on the back, at least for that specific number. Sure the wind may be blowing at 1.2 kph right now and you're worried you're not optimized, but in 30 seconds the wind's gonna change to 2.0 anyway. E.G. tiny optimizations will be nothing compared to the flow of meta. Someone could post a lobster meme thread tomorrow, boost amph usage across the board 30%, even on maps amph clearly shouldn't be used on (although I'm curious about that Amph-stealth-OP hypothesis :D ), and consequently see amph win ratio decrease 7% for a month.

But, the juiciest part: biases that could affect the data that we have here. For instance, let's say people like going gunships for fun, you'll see them played even on maps that they might be only a suboptimal 40% on. If you buff gunships to compensate for these win reductions, then there might be even worse maps that people are now willing to have fun with gunships on, and create a self-reinforcing balance nightmare. And the opposite goes for facs people don't prefer.

Was gonna add more, will maybe do so later.
+2 / -0

12 days ago
Is there a way to compound the data with win probability based on level difference? For example, if a victory was 40% probable according to the ladder, it counts as 0.8 victory instead of 1 (or some other number smaller than 1, not sure what function should be used)
+4 / -0

11 days ago
Shield 60% win-rate at 2000+?
+0 / -0
Page of 3 (59 records)