|
![[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
|
Canada8159 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.
|
On July 28 2014 07:09 KingAlphard wrote:
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.
Agreed. If you want to see bigger buckets, they're actually already on Spawning Tool
PvT
PvZ
TvZ
I made this particular graph because of a request from another member of the community.
Any thoughts out of that data? The trends there are particularly dramatic, I think.
|
This is pretty awesome. Shows pretty much where all my losses happen. Not surprised.
|
On July 28 2014 07:49 DeathSoror wrote:Show nested quote +On July 28 2014 07:09 KingAlphard wrote:
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. Agreed. If you want to see bigger buckets, they're actually already on Spawning Tool PvTPvZTvZI made this particular graph because of a request from another member of the community. Any thoughts out of that data? The trends there are particularly dramatic, I think.
I don't know about TvZ but the winrates in protoss matchups seem pretty accurate. For example the PvZ one clearly shows the strength of protoss 2 base all ins (64% winrate between 8-12 minutes)
|
Interesting. Thanks.
Using 60% for the threshold for "very strong,"
P is very strong vs T: <10.5 mins (toss allins and doomed terran trickiness), and for blips >20 mins (tech related timings and max-outs?)
P is very strong vs Z: 9-12 mins, typical allin timings
T is very strong vs P: never ??
T is very strong vs Z: <5 mins (rax allins) and for blips >20 mins (when zerg loses his broods from no ground support?)
Z is very strong vs T: 6 mins (failed rax allins), 7.6-10 mins (zerg allins), and then never again
Z is very strong vs P: <7.5 mins (failed cannon rushes?), mostly and 2 later blips
......................
Sort what we all know. 3 comments:
No wonder people hate protoss: they excel in the allin phase, and then are very competent in the lategame. Gives the impression that they combine trickiness, cheese and potential power which "appears" OP.
No wonder terrans complain: The only time terrans really seem to shine is with cheesy barracks pressure against zergs. The early game is extremely hazardous and defensive. Whats most surprising is how small the terran midgame advantage actually is in terms of numbers, compared to how strong the other races can be in certain game phases. Its longer than it is large. I think the intention of recent patches is to give terran a bigger midgame advantage, so they can be "very strong" somewhere besides <5 mins.
In midgame-length macro games, no race seems to be extremely advantaged, though a bunch of timings seem to pop up. Again, I think the patches are intended to sort of give terran the advantage in the midgame macro games which is interesting. To get a big edge, it seems like we should allin more with Z and P, and probably cannon rush less (unless you're weedamins)! Incidentally, they say the Korean server is way cheesier than ours. They must have figured this out years ago!
|
Is there a way to check population sizes per race in this data?
|
Would be cool if you draw a thick line at 50% and a thin line at every other 10%.
|
On July 28 2014 09:01 pure.Wasted wrote:
Is there a way to check population sizes per race in this data?
I posted on reddit a few months ago about win rates based on supply difference. The original post is here. The specific graphs are at
http://i.imgur.com/IvlqqIG.jpg
http://i.imgur.com/J27UQTg.jpg
Is that what you were looking for? If not, can you maybe explain a bit more about what you were thinking of?
|
On July 28 2014 09:19 DeathSoror wrote:Show nested quote +On July 28 2014 09:01 pure.Wasted wrote:
Is there a way to check population sizes per race in this data? I posted on reddit a few months ago about win rates based on supply difference. The original post is here. The specific graphs are at http://i.imgur.com/IvlqqIG.jpghttp://i.imgur.com/J27UQTg.jpgIs that what you were looking for? If not, can you maybe explain a bit more about what you were thinking of?
My apologies, didn't realize there was room for misinterpretation! What I meant was how many players there are per race in this data. Because if it's not approximately 33-33-33, then this throws a pretty huge wrench into balance/design assessments.
|
You can tell there aren't enough games in this statistic because of how sharp some of the turns are (is it really this big a change to be at 28min or 29min in TvZ? Doubtful). Still it's a valuable effort, if there was a way to add a lot more matches to it the result would be interesting (or at least, even more than it is now)
|
My speculations:
![[image loading]](http://i.imgur.com/R0vmXjS.png)
|
On July 28 2014 09:22 pure.Wasted wrote:Show nested quote +On July 28 2014 09:19 DeathSoror wrote:On July 28 2014 09:01 pure.Wasted wrote:
Is there a way to check population sizes per race in this data? I posted on reddit a few months ago about win rates based on supply difference. The original post is here. The specific graphs are at http://i.imgur.com/IvlqqIG.jpghttp://i.imgur.com/J27UQTg.jpgIs that what you were looking for? If not, can you maybe explain a bit more about what you were thinking of? My apologies, didn't realize there was room for misinterpretation! What I meant was how many players there are per race in this data. Because if it's not approximately 33-33-33, then this throws a pretty huge wrench into balance/design assessments.
Oh, got it. My bad: I was totally thinking about this in the wrong context.
That's a hard question to answer because of the prevalence of smurf accounts and multi-region accounts. I try to normalize for htat, but one can't be sure.
The best guess I have is 110 Protoss, 78 Terran, and 104 Zerg. Worth noting that even amongst them, there's huge variation in the number of games they have each played. I'm guessing that actually represents each race's prevalence in top-tier competitions?
|
Using the data from the reddit post, I made some graphs about when different races win, if they win.
This doesn't really say anything about balance, but it's rather interesting (to me, at least) to see at what points in the game the different races are (or were, the data covers a fair bit of time) strong or weak.
Imgur album
|
On July 28 2014 06:36 Terranist wrote:Show nested quote +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.
That isn't what it says. It says if the game ends at 8 minutes, there is an 80% chance the Protoss won. Let's say a Dark Templar rush has a 20% chance to outright kill Terran at 8 minutes, and it is the only build that ends the game around that time. Well, then Protoss is going to have a great showing at 8 minutes, because if the game happens to end then, it is likely Protoss won with DT rush. But the other 80% of the time when Terran holds the all-in and the game goes longer, well then Terran wins at X time (which in turn, skews Terran win rates at X time).
So this statistic is actually useless and misleading. Opening with a 6 pool probably has a terrible winrate vs Terran, but if the game ends at 5 minutes, Zerg probably won with it, unless it went completely terribly and Zerg can't transition and quit (hence the win rate is ~60% at that time). But if the Zerg stays a bit longer after failing with the 6 pool and plays from behind... then the Terran gets the win at some other minute mark.
To make this statistic significant, we need to know what the players were doing... this thread should be closed to avoid spreading misinformation regarding balance.
|
I think this would be nice side-by-side with a number of victories over time chart for each race in each non-mirror matchup. So you could see if Terran has X wins at the 5 minute mark, and then Y wins at the 10 minute mark, it would reveal more about the volume of games played/won at each time.
|
|
|
|
|
|
EPT
