|
![[image loading]](http://i.imgur.com/XaqXEDA.png)
Here's the win rates in the non-mirror matchups by game length, which roughly looks at what points in the game certain races appear to have an advantage.
The data comes from roughly 2000-3000 games in each matchup from roughly April 2013 to present on Spawning Tool. The vast majority fo the replays are from tournament replay packs and should represent high-level play, though other games are not filtered out. Note that this does integrate data over multiple balance patches (I'm happy to rerun the numbers within specific time frames upon request).
There aren't many games shorter than 5 minutes or longer than 31 minutes so the graph is truncated there. On the shorter end, those are ignored since many of them appear to be re-games, and on the longer end, those are grouped together. The raw data is available here.
I'm not really sure how to read the data myself. There are definitely deviations from 50-50 along the way, though it's heartening to know that it's pretty close in all matchups in traditional late-game timings (19-25 minutes). It's a little crazy between 25-31, but that might be small sample sizes. The sample for >31 minutes is pretty big, though, and that is very close for 2 of 3 matchups.
Were I a better person, I would have rendered this online to make the graph slighly more interactive, but as it is, I unfortunately am not. If you would like more interactive visualizations, however, chime back, and I'll put more effort into that in the future. If you have any other thoughts on other graphs or data you would like to see, I'm happy to take all suggestions for that as well.
Also happy to get help on doing some of this. Someone recommended error bars on the last one, and I actually don't know what the right methodology and presentation for that is. More generally, I could use help on methodology and presentation, so call me out on anything.
|
This is awesome! I like the idea of making one graph for each major patch, and then updating these every several months as part of an ongoing occasional series.
|
I like these kind of stats, a matchup could be 50/50 but still imbalanced, so the biggest spike is 80% win rate protoss when the game end at 7-8min its interesting but for each spike there is a counter spike. Because if this push dont work terran are probably ahead
|
Canada8160 Posts
Why games from over a year ago and not more recently?
|
On July 28 2014 05:28 Pontius Pirate wrote:
This is awesome! I like the idea of making one graph for each major patch, and then updating these every several months as part of an ongoing occasional series.
I wish I could do it patch-by-patch! Unfortunately, I just don't get enough data for it. Since I basically only use replay packs from professional tournaments, I actually only get a few hundred games for each patch, which isn't nearly enough data to get representative samples for graphs like this. I guess that's just another reason to appreciate the organizers who do release replay packs for us!
|
In this thread: proof that early game all-ins are Protoss's strength
|
On July 28 2014 05:47 User15937 wrote:
In this thread: proof that early game all-ins are Protoss's strength
Zerg all-ins are pretty good, too.
|
Someone needs to see this and you know who I am talking about.
|
time for that 27 minute mark timings, zergs.
|
I think this is a metric that blizzard can use more of to determine "relative balance".
|
On July 28 2014 06:12 Phanekim wrote:
I think this is a metric that blizzard can use more of to determine "relative balance".
Blizzard has access to every replay in existence, not just certain specific tournaments, and David Kim has stated that they have specialized analytic tools with which to break them down. I'm quite sure they already have charts like these at the HQ that are even more sophisticated. It's still awesome to get to see something similar as a non blizzard employee.
|
|
|
On July 28 2014 06:17 SatedSC2 wrote:
Glad it's not just me who finds that PvT mid-game difficult T_T
in all fairness protoss does have an 80% winrate at 8 minutes in PvT.
|
On July 28 2014 06:17 SatedSC2 wrote:
Glad it's not just me who finds that PvT mid-game difficult T_T
It's not super difficult, just a 8 minute time interval where you lose 60% of the games.
It's interesting to see that T early game is the worst (say the first 10 minutes, or before first medivacs come out). while i would assume T late game would be an important losing point for most T.
|
It should be pointed out that a high win% at a certain time is really more of an indicator of the prevalence of certain strategies, instead of being a commentary on balance. For example, if a similar graph were constructed for TvT for just bio vs mech play we might imagine that the graph would show that the bio player win% is highest at the beginning stages of the game and into the early mid-game and finally with the mech player win% overtaking the bio player's as the time-elapsed moves into the late game.
This doesn't necessarily mean that mech play is harder in the early game and easy in the late game. It's just an acknowledgment of the fact that bio players would have more ways to kill a mech player in the early game. In order to get a better idea on balance you would probably also have to add in a bell graph in the background showing the distribution of game lengths for each MU (for example if we knew that 90% of the games ended in bio vs. mech before the lategame, we could draw better conclusions about the viability of mech as a strategy.)
|
On July 28 2014 06:38 YumYumGranola wrote:
It should be pointed out that a high win% at a certain time is really more of an indicator of the prevalence of certain strategies, instead of being a commentary on balance. For example, if a similar graph were constructed for TvT for just bio vs mech play we might imagine that the graph would show that the bio player win% is highest at the beginning stages of the game and into the early mid-game and finally with the mech player win% overtaking the bio player's as the time-elapsed moves into the late game.
This doesn't necessarily mean that mech play is harder in the early game and easy in the late game. It's just an acknowledgment of the fact that bio players would have more ways to kill a mech player in the early game. In order to get a better idea on balance you would probably also have to add in a bell graph in the background showing the distribution of game lengths for each MU (for example if we knew that 90% of the games ended in bio vs. mech before the lategame, we could draw better conclusions about the viability of mech as a strategy.)
This is a good point! We don't know how many games have ended in the early, mid or endgame.
I assume that the games are all summed up per minute and the win rates are calculated per minute, so this means a bell graph per minute would be interesting as well (or a total number of games per minute per match up).
|
On July 28 2014 05:52 CakeSauc3 wrote:Show nested quote +On July 28 2014 05:47 User15937 wrote:
In this thread: proof that early game all-ins are Protoss's strength Zerg all-ins are pretty good, too.
11/11 too when it's not failing haha
|
Looking at the graph you can see that it changes too much between 1 minute intervals. That's because there aren't enough data.
The most reliable part is probably the early game where you can see which race has the better all ins.
PvT = Protoss obviously
TvZ = Terran at <5 minutes (11-11) then Zerg
PvZ = Zerg (early pools)
IMO you should use bigger intervals where you have less data. For example instead of 26-27 , 27-28 minutes get 25-30, then maybe 30-40, etc.
|
On July 28 2014 05:52 CakeSauc3 wrote:Show nested quote +On July 28 2014 05:47 User15937 wrote:
In this thread: proof that early game all-ins are Protoss's strength Zerg all-ins are pretty good, too.
All-ins are good for each race
|
It seems that terran has better mid-game all-ins in PvT. The protoss early advantage could be early oracle rush related, no? Then, in PvZ, zerg has the advantage early because of early pools, then protoss takes the advantage when all the all-ins come into play.
|
|
|
|
|
|
EPT
