|
On May 03 2012 23:12 Telenil wrote:
Edit:
By the way, statistics question: I guess the "global" results have smaller error bars because their are more games involved,but since these global results are made from each of the separate matchups below, wouldn't it make more sense to sum the errors quadratically instead? Like, if you have a 3% uncertainty on TvZ and a 3% uncertainty on TvP, claiming there is a 2% uncertainty on the global T win rate looks horribly wrong from my physicist perspective.
You sum the absolute errors quadratically. The graph shows the relative errors, since winrates are just the fraction of games won. The error on winrates is the relative error on the number of games won. If you add up data from 2 independent sources, you expect the relative error to decrease, which is what happens.
|
Sample size of Korea is laughable. Tells more about some player's personal skill, nothing about balance.
|
For comparison, here are Brood War winrates from 2007-2011.
+ Show Spoiler +
Look at ZvP, where Z is favored for 2 years straight, or how Protoss is the lowest race from 2008-2010! Nothing in the April winrates for SC2 worries me too much.
|
International is so balanced, holy shit!
|
On May 03 2012 23:17 Equity213 wrote:
wtf korean terrans < 50%
that just made my day :D
Finally protoss showing some strength, so awsome.
I wouldn't get the champagne out yet. Korean sample size is small and the win rates there are absurdly volatile. Compare the PvT matchup between March and April.
Actually what I did find interesting is how those months repeat what we saw in October/November of last year almost exactly. Specifically a massive and statistically significant imbalance (as demonstrated by the non-overlapping error bars) in favour of Terran swinging to a non-significant increase in favour of Protoss the next month. I wonder if that could be significant somehow...did something happen in Korea in March/April that also happened October/November?
|
On May 03 2012 23:22 Rannasha wrote:Show nested quote +On May 03 2012 23:12 Telenil wrote:
Edit:
By the way, statistics question: I guess the "global" results have smaller error bars because their are more games involved,but since these global results are made from each of the separate matchups below, wouldn't it make more sense to sum the errors quadratically instead? Like, if you have a 3% uncertainty on TvZ and a 3% uncertainty on TvP, claiming there is a 2% uncertainty on the global T win rate looks horribly wrong from my physicist perspective. You sum the absolute errors quadratically. The graph shows the relative errors, since winrates are just the fraction of games won. The error on winrates is the relative error on the number of games won. If you add up data from 2 independent sources, you expect the relative error to decrease, which is what happens. Here the two sources measure different things, don't they?
Hm, I guess it's just the "global ratio" that is meaningless as far as balance is concerned. Indeed a race could win 60%+-2 against the 2nd and 40%+-2 against the third, and have a 50%+-1 global ratio, it's just that the 50+-1 would have no significance as far as balance is concerned.
That's probably what confused me in the first place, thanks for the clarification.On May 03 2012 23:39 Lightspeaker wrote:Show nested quote +On May 03 2012 23:17 Equity213 wrote:
wtf korean terrans < 50%
that just made my day :D
Finally protoss showing some strength, so awsome. I wouldn't get the champagne out yet. Korean sample size is small and the win rates there are absurdly volatile. Compare the PvT matchup between March and April. Actually what I did find interesting is how those months repeat what we saw in October/November of last year almost exactly. Specifically a massive and statistically significant imbalance (as demonstrated by the non-overlapping error bars) in favour of Terran swinging to a non-significant increase in favour of Protoss the next month. I wonder if that could be significant somehow...did something happen in Korea in March/April that also happened October/November?
Perhaps a specific player trampled everyone else in the tournaments? The error bars are calculated using the binomial distribution, so it assumes the outcome of the matches can be modeled as random. But it is not necessarily true at such a small scale: a player performing exceptionally well could shift the graph by 5% or more.
|
On May 03 2012 23:56 Telenil wrote:Show nested quote +On May 03 2012 23:22 Rannasha wrote:On May 03 2012 23:12 Telenil wrote:
Edit:
By the way, statistics question: I guess the "global" results have smaller error bars because their are more games involved,but since these global results are made from each of the separate matchups below, wouldn't it make more sense to sum the errors quadratically instead? Like, if you have a 3% uncertainty on TvZ and a 3% uncertainty on TvP, claiming there is a 2% uncertainty on the global T win rate looks horribly wrong from my physicist perspective. You sum the absolute errors quadratically. The graph shows the relative errors, since winrates are just the fraction of games won. The error on winrates is the relative error on the number of games won. If you add up data from 2 independent sources, you expect the relative error to decrease, which is what happens. Here the two sources measure different things, don't they? Hm, I guess it's just the "global ratio" that is meaningless as far as balance is concerned. Indeed a race could win 60%+-2 against the 2nd and 40%+-2 against the third, and have a 50%+-1 global ratio, it's just that the 50+-1 would have no significance as far as balance is concerned. That's probably what confused me in the first place, thanks for the clarification.
Yeah, the global ratio is nice and all, but it's not nearly as interesting as the individual matchup figures.
|
" ZvT 49,9% - 50,1%"
NO wonder I can't win in this mu! 
On May 03 2012 23:33 SarcasmMonster wrote:For comparison, here are Brood War winrates from 2007-2011. + Show Spoiler +Look at ZvP, where Z is favored for 2 years straight, or how Protoss is the lowest race from 2008-2010! Nothing in the April winrates for SC2 worries me too much.
Yeah really interesting to look at. SC2's graphs almost seems better in comparison 
|
On May 04 2012 00:35 Cereb wrote:" ZvT 49,9% - 50,1%" NO wonder I can't win in this mu! Show nested quote +On May 03 2012 23:33 SarcasmMonster wrote:For comparison, here are Brood War winrates from 2007-2011. + Show Spoiler +Look at ZvP, where Z is favored for 2 years straight, or how Protoss is the lowest race from 2008-2010! Nothing in the April winrates for SC2 worries me too much. Yeah really interesting to look at. SC2's graphs almost seems better in comparison
Kind of wished the graph tracked Brood War stats in its early years to see what crazy shenanigans occurred in its infancy as a e-sport. But yeah, the SC2 graph isn't all too bad.
|
On May 03 2012 23:33 SarcasmMonster wrote:For comparison, here are Brood War winrates from 2007-2011. + Show Spoiler +Look at ZvP, where Z is favored for 2 years straight, or how Protoss is the lowest race from 2008-2010! Nothing in the April winrates for SC2 worries me too much.
Interestingly enough, that's also the period where my ZvP winrate started going downhill... Seems like i've never liked the way zvp is supposed to be played with all this 3 bases stuff and evolution chamber to wall and sh.t... Much like in sc2... Why is that ? T_______________T
|
On May 03 2012 21:39 ZenithM wrote:
It may be the first time in the history of the game that Korean Terran have less than 50% winrate overall.
How the mighty have fallen.
Expect this to fall more. Even internationally, win rates have failed to dip below 50% for the reasons below. However, they will also follow the Korean trend.
-----
Rank 1 master Terran here and I'm astounded DK would not address Late game TvP. I've kept track of my win % on sc2gears and passed 20 in-game minutes -- I have only won a single TvP this entire season.
Given this disparity on an anecdotal level (my own), it may not have much significance. However, coupled with many other Rank 1-8 Master Terrans who also fail to win passed 20 in-game minutes, and have tracked a similar rate of failure on sc2gears, I'm guessing that this data exists within the released win rates per MU at least on our level, if not elsewhere.
A few explanations of the win rates: Many of the weaker Terran players have left the race entirely and switched to Protoss or Zerg. The evidence of this is in the proportion of Terran players to their Zerg and Protoss counterparts in the GM and Master leagues within NA and EU. Terran players have dropped to 1/4 of the race selections. (In KR, the Korean Terrans are on another level, and yet their representation is only equal).
Note that it is a possibility that the win rate of the remaining Terrans never changed, but the overall win rate of Terran simply increased from the reduction of the weaker Terran players.
Thus, this is one explanation of relatively even win rates TvP.
An alternative explanation of these win rates is that Terrans are opting for more All-in or Semi all-in mid game timings. Most good Terran players recognize that our Win % peters out as the game goes on, and so we go for timings when we have the greatest chance to win.
A more thorough discussion on balance would occur when Win % of Races were determined per unit of time (0-3 minutes, 3-6 minutes, etc.) and also by League and Server.
I feel that Terran has been pigeonholed into the Terran that we have today. All races should be able to perform relatively evenly at all points in the game, and the winner should be determined by the criterion of his skill, not overwhelmingly by the strength of a race at a certain point in time. If Blizzard finds that Terran has a disproportionate win rate from the 6-9 minute mark, but also a disproportionate loss ratio passed 20 minutes, then there should be adjustments made to weaken Terran at the 6-9 minute mark and strengthen Terran's late game options. I'll take another Rax/add-on timing nerf (given all of the implications) if it means Ravens get faster movement and HSM speed and something is done with BCs.
|
Why not give us winrates for certain time periods in games...? Would be much more relevant and useful for balancing. Terran early game cheese and protoss late game are both too strong.
|
Can someone explain the numbers to me? Like it says for korea that 157 games were played, but then right under it for zvp there are 1207 games.. so there were more games played for 1 matchup then there were for all of them?....I don't get it
|
First time ever korean terran is the weakest race. IS THIS HEAVEN?
|
On May 03 2012 21:40 Raid wrote:
Can terrans use the same arguments that zerg and protoss used of that we should "balance the game at the highest level" and since terran doing so bad in korea they need buff?
No they can't since they're not doing bad in korea, for the first time since July there is only one terran in the semi finals of Code S.
It is also for the first time since the game came out that they are beneath an overall winratio of 50%.
Terran isn't doing bad, zerg isn't really underperforming either, it's just a really good protoss season and we're gonna have to wait and see.
What's more important is the number of games is lower than usual with only 157 from the three matchups.
Can someone explain the numbers to me? Like it says for korea that 157 games were played, but then right under it for zvp there are 1207 games.. so there were more games played for 1 matchup then there were for all of them?....I don't get it
That's the number of games played in ZvP total since May 2011, 157 refers to number of games for all matchups in April.
|
Vatican City State582 Posts
korean terrans < 50% win rate
kind of sad, but the differential isn't that bad, tbh
|
Korean Terran being at less than 50% winrate feel wrong.
Most of the time they are the better players.. meh.
Will see next month.
I'm sad Korea is always such a small amount of games.
|
It might be true that the Korean sample size is small, but it's also where the highest level of skill is, and that's where the balance should focus upon. I mean, look at the EU protosses. None has had really good results, only mediocre. But majority of top Korean protosses are doing soo well. EU pros need to see why that's happening, and try to translate that to good results
|
Come on MVP! Do it for the Aeon of Strife!
I know some hate MVP for cheesing, but having a sole Terran hope is what we all need. THe downside is that if he wins the GSL, people will just say Terran is still imba, sort of like how everyone said ZvP was still imba since DRG won the last GSL vs Genius.
|
Just look at how wildly TvP swung around from March to April => it's just a huge meta game shift. I'm sure T will figure out a new style soon...Stuff like this has happened often and will happen often, it's not as big as people make it out to be, don't worry. That said, I hope the P favored meta persists until the GSL finals, I want to see a PvP final for once
|
|
|
|
|
|
EPT![[image loading]](http://i.imgur.com/gmXwO.png)
