|
Should Terran 2rax all day, every day? Should Zerg play for the late game in every matchup? Is TvP a race against the clock for Terran, losing whenever the Protoss gets Colossus and Templar Tech?
Using SC2gears and some Excel kung-fu, I tried to answer these questions and more by looking at race win rates based on the length of the game from all games at MLG Providence. I came up with this set of charts showing what happens as the game progresses (click on the image for all the charts).
![[image loading]](http://i.imgur.com/8Wfcc.png)
Data here: MLG Providence Race/Length of Game Win Rates Note: times here use in-game time, not wall clock time. 5 minutes of wall-clock time is about 7 minutes of game-time
The data does show some interesting results. But I'm not satisfied that the sample size is large enough. I'd love to repeat this analysis on a bigger dataset if there is fresher and larger replay pack lying around (coughplayhemcough).
To the Pros, Masters players, and everyone: do the numbers here line up with how you think about the game? Is Terran overmatched in the late game? Does Zerg grow stronger as the game goes on? Should you 6-pool (100% Z win rate in PvZ <5 minute games) or 2-rax every (80% T win rate in ZvT 5-10 minutes) every game? Discuss.
|
I like threads and graphs like these, but always have something to say about the data (don't stop though, for sure).
-Providence: not all matchups really even in skill :p
-On average more >20min TvZ, ZvP and TvP than TvT matchups (maybe the TvTs were like 40 minutes long, but add em up guys) - maybe there were just fewer mirrors in general though. Actually that sounds right.
But I'm not satisfied that the sample size is large enough
Agreed
Also looking at the ridiculous increase in Z and decrease for P before 5 minutes, I'm guessing 6pool is the build against P 
|
On February 05 2012 06:32 Ktk wrote:
-On average more >20min TvZ, ZvP and TvP than TvT matchups (maybe the TvTs were like 40 minutes long, but add em up guys) - maybe there were just fewer mirrors in general though. Actually that sounds right.
I stopped at 30 minutes, and there the most common MU was TvT, TvZ was second and either ZvZ or ZvP third.
On February 05 2012 06:32 Ktk wrote:Also looking at the ridiculous increase in Z and decrease for P before 5 minutes, I'm guessing 6pool is the build against P
Yes, this represents 3 successful 6-pools, in the same way most of the 5-10 minute TvZ games are successful 2rax.
|
Hahahaha, ZvZ oh you...
BTW by all MLG games you mean even open bracket?
Interesting data, don´t know if its really that relevant but its always interesting to see stuff like this
|
I thing line graphs are not the right ones to be used in this kind of analysis, specially at the end of the first one, which goes from 20+ to overall.
|
On February 05 2012 06:47 windsupernova wrote:
Hahahaha, ZvZ oh you...
BTW by all MLG games you mean even open bracket?
Interesting data, don´t know if its really that relevant but its always interesting to see stuff like this
Yes this includes the open bracket. If I used only pool play and the Sunday bracket there wouldn't be a meaningful number of games.
On February 05 2012 06:47 Hoon wrote:
I thing line graphs are not the right ones to be used in this kind of analysis, specially at the end of the first one, which goes from 20+ to overall.
Yeah ... what I wanted to do was just draw a straight line representing overall WRs. If I get a bigger dataset I'll try to make better graphs.
|
To me this is how balance stats should be presented, the length of game variable is really something that can say a lot. Hopefully you get a bigger sample next time =D
|
20+ is very inaccurate :s Any chance you could do it until you reach 40+?
|
I guess it makes somewhat intuitive sense, but it'd be nice to have a bigger sample size.
|
Wow, nice graphs, also it is about time someone did something like this.
I just hope you manage to get bigger sample next time. I'll definitely look forward for it. =)
|
On February 05 2012 07:14 decaf wrote:
20+ is very inaccurate :s Any chance you could do it until you reach 40+?
I'd like to, but the sample size gets too small even at 30 minutes. By setting the cap at 20 minutes you get a sample that's roughly the size of the previous increments. I would need a much larger data set to go past 20 minutes.
Also by 20 minutes unless something has gone wrong, the game really ought to be in the "late game". Players' gas income is high enough to support lots of high-tier units. Neither player is really building up their economy much more. Most upgrades & tech has been completed. The game is much more about protecting later expansions, minimizing harassment damage, trading armies cost effectively, etc., than it is about exploiting timing attacks.
|
Very good indication of the strengths of each race in my opinion. Thanks for this!
|
Ha, love how winrates for Zerg is so high below 5 minutes. The good old 6 pool.
|
Really awesome stats.
Thanks for putting this together. 
|
late game T looks nice O_o
thx for the graph man !
|
On February 05 2012 07:24 Bagration wrote:
Ha, love how winrates for Zerg is so high below 5 minutes. The good old 6 pool.
still the best cheese, screw cannons. it's the mother of cheese. you can call it milk. or even cow.
|
With sc2gear can we know the pourcentage of win of a strategy? Can you know the unit composition at the end of each game?
|
The 'Overall' at the end of the first graph kinda gives the wrong impression. The X-axis there represents time which is increasing left-to-right, while 'Overall' is simply the overall winrate, not based on time. If someone doesn't read the X-axis it seems like Terran winrate steadily goes down as time progresses, then suddenly spikes at the end.
But other than that nice data.
|
On February 05 2012 07:24 K3Nyy wrote:
Very good indication of the strengths of each race in my opinion. Thanks for this!
or current metagame
|
This is what i would expect. It feels like im on the clock vs zerg and protoss when i play as terran
|
|
|
|
|
|
EPT
