EPTWould have thought TvP to be quite off after the 60-65% Terran winrate in GSL. Seems like it's just a GSL-slump and Protoss are doing well everywhere else.
April win percentages - Page 7
| Forum Index > SC2 General |
|
Big J
Austria16289 Posts
Would have thought TvP to be quite off after the 60-65% Terran winrate in GSL. Seems like it's just a GSL-slump and Protoss are doing well everywhere else. | ||
|
SniXSniPe
United States1938 Posts
direct link: https://docs.google.com/file/d/0B76Yjrn8DAmfeWdpY0F0b1ByOEU/edit You'll notice some differences. | ||
|
Fingerpin
Denmark10 Posts
Zerg isn't as bad as many people complain the race to be. | ||
|
Eventine
United States307 Posts
https://docs.google.com/spreadsheet/ccc?key=0Aqh9-jZtcUrrdFFoMDdqal9GdHdrQ05tbm9nbFUyVlE#gid=0 TvP Overall T wins 49.28% of the time z = -1.08, fail to reject null that this is same as 50% TvP Korean Only T wins 43.02% of the time z = -23.02, reject the null Note that this is heavily influenced by GSL qualifiers where T loses far more often, take this out of the equation, you'd get a similar a fail to reject the null PvZ Overall Zerg wins 47.93% oft the time, z = -3.41, you can reject the null here PvZ Korean Only Zerg wins 49.46% of the time, z = -1.72, fail to reject the null | ||
|
Eventine
United States307 Posts
On May 03 2013 05:11 SniXSniPe wrote: http://www.reddit.com/r/starcraft/comments/1dkmgt/hots_korea_only_stats/ direct link: https://docs.google.com/file/d/0B76Yjrn8DAmfeWdpY0F0b1ByOEU/edit You'll notice some differences. I updated my spreadsheet with that information (sheet 4), you still see significant results. But TvZ is closer than the other data sample. It does bring into question the quality of data here. | ||
|
DemigodcelpH
1138 Posts
On May 03 2013 00:24 FakeDeath wrote: And SH needs a buff in burrow-reburrow time and slight movement speed increase. The Swarmhost absolutely does not need a buff in any shape or form if you've seen it used properly. | ||
|
EleanorRIgby
Canada3923 Posts
On May 03 2013 05:26 DemigodcelpH wrote: The Swarmhost absolutely does not need a buff in any shape or form if you've seen it used properly. yea in the burrow times are buffed you would never be able to get to them lol, with proper SH micro you can retreat them pretty fast while spawning scarabs. | ||
|
[OGN]Remmy
United States1206 Posts
| ||
|
Sissors
1395 Posts
On May 03 2013 05:14 Eventine wrote: Because I was curious, I did the same analysis on the other races... splitting each time, overall vs korean tournaments only. It's all in the spreadsheet. It matters which percentage you are using to decide if you reject the null hypophesis. But the problem remains it is hard to say if you got enough data. Sure if it would be a nice binomial distribution it would be easy to calculate and you would have definately enough, but it isn't a binomial distribution, and as shown that just deleting WCS quals made an enormous difference, I think for sure right now there is not enough data to just use statistics to proof possible imbalances. | ||
|
Eventine
United States307 Posts
On May 03 2013 05:49 Sissors wrote: It matters which percentage you are using to decide if you reject the null hypophesis. However the main problem is that you are testing it is a binominal distribution with a certain chance (50%), and that isn't the case, and will never be the case. Simply because that would also assume that each player is equal: So it would mean that Life has an equal chance on a win with Zerg vs Innovation as the chance he beats Snute, which obviously is not the case: For a binominal distribution the chance for each 'experiment' should be equal, which is not the case. And there you get the question how significant the data is: if it was a binominal distribution it would be significant, no doubt whatsoever about it. But it isn't, so you need enough games to make sure you can approximate it with a binominal distribution, which quite obviously isn't the case yet, otherwise just deleting WCS quals shouldn't have such a large impact. Edit: at least I assumed you used a binominal distribution, but not sure after looking at the sheets  hm, you're right, I didn't use binomial distribution. That is probably the wrong method I used, I was doing it too quickly without thinking. i need to think on this as it's been a while since i've looked into binomial distributions. i'm making the assumption that aggregating the data together will take out the influence of skill on the matchup. we have no reason to believe that all the "skilled players" play terran and the "lesser skilled" players play zerg. instead, i'm assuming that there's an equal balance of skill within the races and within the games played. I think that's a fair assumption as we get into larger sample size I think this is especially true when we start restricting the sample to only the koreans leagues where theoretically the best players play. there will still be variances in player skills, but honestly i think high level results are enough, if someone wants to make this a dissertation, go forth and conquer. | ||
|
Sissors
1395 Posts
| ||
|
Sandermatt
Switzerland1365 Posts
On May 03 2013 06:04 Eventine wrote: hm, you're right, I didn't use binomial distribution. That is probably the wrong method I used, I was doing it too quickly without thinking. i need to think on this as it's been a while since i've looked into binomial distributions. i'm making the assumption that aggregating the data together will take out the influence of skill on the matchup. we have no reason to believe that all the "skilled players" play terran and the "lesser skilled" players play zerg. instead, i'm assuming that there's an equal balance of skill within the races and within the games played. I think that's a fair assumption as we get into larger sample size I think this is especially true when we start restricting the sample to only the koreans leagues where theoretically the best players play. there will still be variances in player skills, but honestly i think high level results are enough, if someone wants to make this a dissertation, go forth and conquer. Yes all races are equally skilled, but not all players are equally skilled. If you have 100 matches of which 50 are absolutely loopsided, this means your statistics appears more reliable than it is. Of course this factor cannot be calculated and we have to ignore it for making the calculations. But just keep in mind, that the confidence inerval of the final result is probably slightly larger than what it appears to be. Also correlated results (same players playing multiple games) increase this trend. Go for a bit a larger standard deviation until you say it is statistically significant than you would if you had statistically perfect data. | ||
|
Eventine
United States307 Posts
On May 03 2013 06:17 Sissors wrote: I agree it is a fair assumption with a larger sample size, but the question is when do we get to the larger sample size exactly. We do have a few hunderd games for each match-up, but it is a not a few hunderd uncorrelated games. The winner of an average tournament plays quite some games, which are all correlated games, so they don't count as much as uncorrelated games would count. But well I leave it for math students to look at that, I always disliked statistics, and luckily for my work I can just stick to normal distributions Very true. The winning player does play far more games and has a far greater influence on the match up results. Not much we can do about this except try to wait and gather more data, or perhaps we go down towards leagues that are a bit more skill equivalent, maybe Code A only. On May 03 2013 06:20 Sandermatt wrote: Yes all races are equally skilled, but not all players are equally skilled. If you have 100 matches of which 50 are absolutely loopsided, this means your statistics appears more reliable than it is. Of course this factor cannot be calculated and we have to ignore it for making the calculations. But just keep in mind, that the confidence inerval of the final result is probably slightly larger than what it appears to be. Also correlated results (same players playing multiple games) increase this trend. Go for a bit a larger standard deviation until you say it is statistically significant than you would if you had statistically perfect data. Again as above, we should make the assumption that there is a random distribution of skill among races. Btw, I appreciate all the comments about the statistics, I'm trying to build up better knowledge on my methods so it's a cool exercise. | ||
|
Cloak
United States816 Posts
On May 03 2013 02:12 Emzeeshady wrote: This is terrible logic. Lets say for example there is Race A,B and C which are all perfectly balanced. Lets also say race A is buffed to become much better then Race B and C. Then were three players of equal skill (lets say mid masters) who were all in exactly the same spot. When this buff happens the player of Race A moves far ahead of the other two. Is the race not holding them back? Yes, of course it is possible to move ahead but that would mean they have to also move ahead of the player of Race A in skill to get just as far. I am not saying any of this is the case but saying that balance does not effect non pro players is just ignorant imo. Yea, it's bias, but incompetence is even more bias. There's fewer and fewer ways to do something right, and many more ways to fuck it up. So while there is theoretical balance shift at every player skill (and style and unit composition and stage of the game) it just doesn't seem worthwhile to investigate unless it's otherwise obvious, or it coincides with the more prominent games (pros). | ||
|
Jinky
United States64 Posts
On May 03 2013 00:30 GTPGlitch wrote: @Jinky: That's a really dumb argument, because balance affects players no matter how bad they are. If i'm in bronze, and there's another guy in bronze, and protoss has a 99% winrate vs terran and he's playing protoss, it's very likely he beats me because we're at the same skill level and balance favors the protoss. Sure, I can just get better and win, but throwing the casual players under the bus isn't exactly the way to go about making star2 super popular First of all, there is no 99% win rate in SC2 now or ever in the past. That is an asinine way to argue. But even if there was a 99% win rate, if that win rate includes weaker players then it doesn't matter too much (as far as balance goes). The only win rates that matter are those with players who are so good that they push the limits of what the races can do (ie. top Koreans). Why is this? Because weaker players still aren't hitting their injects or setting up depot wall-offs or placing good force fields, and so they lose because they beat themselves, because they don't use their tools properly. It's only when a race is using all of its tools and still can't win that balance should be talked about. On May 03 2013 02:12 Emzeeshady wrote: This is terrible logic. Lets say for example there is Race A,B and C which are all perfectly balanced. Lets also say race A is buffed to become much better then Race B and C. Then were three players of equal skill (lets say mid masters) who were all in exactly the same spot. When this buff happens the player of Race A moves far ahead of the other two. Is the race not holding them back? Yes, of course it is possible to move ahead but that would mean they have to also move ahead of the player of Race A in skill to get just as far. I am not saying any of this is the case but saying that balance does not effect non pro players is just ignorant imo. In your example, how can you say that the buffed/superior Race A is imbalanced compared to Races B and C even though the best players in the world (top Koreans) are winning vs Race A in a relatively 50:50 ratio? If other human beings are adapting to that "buff" and winning against it, those mid masters in your example should be able to also. Weaker players simply do not matter in terms of balance. The only realm where weaker players (and all players) matter is in terms of fun and desirable gameplay. My point in saying all this is that people need to stop fixating so much on balance and should instead practice more to improve and win. The game is way too complex (the possibilities of unit compositions, timings, army positioning, economic relationships, resource relationships, scouting knowledge or lack thereof, hidden costs, unit value per supply, unit per-hit cost from each specific enemy unit, etc. etc. etc.) for "balance discussions" to revolve around specific units/strategies that seem OP even though there are many other things that can be done to counter these (things that the top Koreans are probably doing, because they are winning versus those units/strategies). The only people that should be concerned with balance are the game developers and the very best players. There were some gaping holes in WoL that I think Blizzard addressed well in HotS. The game is in a much better place than in WoL, and HotS opened many more options for different play styles (thus, more fun for different people). Statistics threads like this don't matter in regards to game balance if the data is not exclusively about top Koreans. | ||
|
guN-viCe
United States687 Posts
..until the nerf stick. | ||
|
LuckyMacro
United States1482 Posts
On May 03 2013 07:49 guN-viCe wrote: Terran is back, baby! ..until the nerf stick. Blizzard already looking at barracks/bunker build time no doubt. | ||
|
Crankyhobo
United States12 Posts
In my opinion zerg got shat on for an expansion that is supposed to be about zerg. | ||
|
Sissors
1395 Posts
On May 03 2013 07:25 Jinky wrote: First of all, there is no 99% win rate in SC2 now or ever in the past. That is an asinine way to argue. But even if there was a 99% win rate, if that win rate includes weaker players then it doesn't matter too much (as far as balance goes). The only win rates that matter are those with players who are so good that they push the limits of what the races can do (ie. top Koreans). Why is this? Because weaker players still aren't hitting their injects or setting up depot wall-offs or placing good force fields, and so they lose because they beat themselves, because they don't use their tools properly. It's only when a race is using all of its tools and still can't win that balance should be talked about. Sorry but thats besides horrible for blizzard sales also just not how it works. If at bronze level an average player who shifts from race X to race Y suddenly alot better, than race Y is too strong at bronze level. Now because of what you describe you can accept higher level of imbalance at bronze level than at code S level. However that doesn't mean it is irrelevant what the balance is at those levels. They are also paying customers, and if they have to play a horribly imbalanced game at their level they soon won't be paying customers anymore (not to mention they also simply deserve a balanced game for them). But to follow your logic, why should we then use the completely arbitrary line of top Koreans for balancing? Why not actually use a situation where those races can actually use all their tools instead of being limitted by our puny humans? So get the top players with best decission making, and let them play via something like automaton 2000 bot. Then the races truly aren't limitted anymore and used properly. And I now know already you aren't in favor of it, apparently you also agree that it should be balanced for regular use by Korean pro's. Then why shouldn't it be balanced for regular use of other players? | ||
|
LeeDawg
United States1306 Posts
terran looked really good in that matchup when the game came out but zergs are adapting, which is a good sign | ||
| ||
