|
On September 02 2014 15:57 Ghanburighan wrote:Show nested quote +On September 02 2014 11:07 sibs wrote:
Terran's doing incredibly overall on aligulac, 58% WR vs Z, 54.5% vs P, 4 out of the 5 top spots, obviously leading race at +6%. I'm actually surprised that P isn't the leading race. The calculation for leading race seems to always have the least intuitive result. As such I wouldn't ever cite the leading race calculation as a balance metric.
The reason you don´t count it is because you don´t agree personally with the result? Or is there some other reason?
|
That sounds like something we could easily prove or disprove. Given how many games we add from early rounds of tournaments or qualifiers nowadays, I tend to think you're wrong there, but I can't be sure. I always wish we could add more.
Well you still have an equal amount of active terrans on the ladder as for the other races. Regarding Aliguac numbers, it's not about terrans dropping out earlier, but rather about fewer terrans pariticpating in the first place. Thus, the average active terran player is much less likely to be considered competitively than the average P/Z-player.
But it's very easy to proof that the combination of actual skills + benefiting from balance matters here. Answer these two questions:
(1) If you play the strongest race are you more likely to be in a higher league than a player who plays the weakest race, ceteris paribius?
(2) Is the average GM player more likely to be paticipate in compeittive play than the average diamond player?
If you answer yes to both questions, it has basically been prooved deductively.
It's ofcourse true that the effect you will see if one race is imbalanced is that it typically will have more players going further in tournaments, and thus more games played. That effect is already taking into account by win/rates. But the effect of the average T player that participates in a tournaments being more skilled than the average P/Z-player is not taken into account in Aliguac numbers, thus some type of adjustments here needs to done.
Thus, it's probably better to use the moving average of games played by race X as that better shows when one race is truly suffering. This is due the logic that it is giving less weight to months where the only reason the representation is changing is due to players of race X getting further in tournaments.
Instead it can only change when there is a consistent reduction in representation by players of race X which can only happen (assuming win/rates do not deviate too far from 50/50) if fewer players of that race participate in the first place.
|
On September 02 2014 15:57 Ghanburighan wrote:Show nested quote +On September 02 2014 11:07 sibs wrote:
Terran's doing incredibly overall on aligulac, 58% WR vs Z, 54.5% vs P, 4 out of the 5 top spots, obviously leading race at +6%. I'm actually surprised that P isn't the leading race. The calculation for leading race seems to always have the least intuitive result. As such I wouldn't ever cite the leading race calculation as a balance metric.
Leading race shows which race is most prominent on top of the board. The issue with this metric is that Aliguac rankings suffer heavily from whether you mostly play vs foreigners or against Kespa koreans. Given that there are more terran players playing outside Korea, it skews the leading race indicator in favor of terran - regardless of balance.
IMO players like Forgg, Polt, MMA, MVP JJaki would never have been on the 1st page if it wasn't for them being on foreign teams. Someone like Yoda is also overrated due to his particpation in tournaments like the Zotac cup.
It's likely a consequence of how well korean terrans performed untill 2012, which increased their brand values, and thus they became more attractive to foreign teams as a result.
Since Aliguac already is aware of the issue, I think they should have been more active in explaining the weakness of relying on certain metrics on their site, which doesn't really say alot about balance, but can confuse a lot of people.
|
On September 02 2014 18:03 Hider wrote:Show nested quote +That sounds like something we could easily prove or disprove. Given how many games we add from early rounds of tournaments or qualifiers nowadays, I tend to think you're wrong there, but I can't be sure. I always wish we could add more. It's ofcourse true that the effect you will see if one race is imbalanced is that it typically will have more players going further in tournaments, and thus more games played. That effect is already taking into account by win/rates. But the effect of the average T player that participates in a tournaments being more skilled than the average P/Z-player is not taken into account in Aliguac numbers, thus some type of adjustments here needs to done.
But the adjustment you're making seems too broad, given that it leads you to conclude there was TvP imbalance at a time where neither the winrates nor the results of the top players support this idea. If your model could include different performance results at different levels of skill, that would be perfect I guess, but that sounds awfully hard to do. The alternative is to be very precise in the commentary of the graph I suppose.
|
On September 02 2014 21:14 Nebuchad wrote:Show nested quote +On September 02 2014 18:03 Hider wrote:That sounds like something we could easily prove or disprove. Given how many games we add from early rounds of tournaments or qualifiers nowadays, I tend to think you're wrong there, but I can't be sure. I always wish we could add more. It's ofcourse true that the effect you will see if one race is imbalanced is that it typically will have more players going further in tournaments, and thus more games played. That effect is already taking into account by win/rates. But the effect of the average T player that participates in a tournaments being more skilled than the average P/Z-player is not taken into account in Aliguac numbers, thus some type of adjustments here needs to done. But the adjustment you're making seems too broad, given that it leads you to conclude there was TvP imbalance at a time where neither the winrates nor the results of the top players support this idea. I f your model could include different performance results at different levels of skill, that would be perfect I guess, but that sounds awfully hard to do. The alternative is to be very precise in the commentary of the graph I suppose.
I agree it would be better, but I guess I am just not that interested in spending time typing in old GSL data. The data I had here for Aliguac was something I set into my model many months ago and I just updated it for August. I posted the graph here mainly as a response to the fact that Aliguac data seems to be the center of many discussions, and I still think this approach I suggested gives a better perspective on balance than naked win/rates. It's definitley not supposed to be the most precise-answer though.
|
On September 02 2014 18:39 Hider wrote:Show nested quote +On September 02 2014 15:57 Ghanburighan wrote:On September 02 2014 11:07 sibs wrote:
Terran's doing incredibly overall on aligulac, 58% WR vs Z, 54.5% vs P, 4 out of the 5 top spots, obviously leading race at +6%. I'm actually surprised that P isn't the leading race. The calculation for leading race seems to always have the least intuitive result. As such I wouldn't ever cite the leading race calculation as a balance metric. Leading race shows which race is most prominent on top of the board. The issue with this metric is that Aliguac rankings suffer heavily from whether you mostly play vs foreigners or against Kespa koreans. Given that there are more terran players playing outside Korea, it skews the leading race indicator in favor of terran - regardless of balance. IMO players like Forgg, Polt, MMA, MVP JJaki would never have been on the 1st page if it wasn't for them being on foreign teams. Someone like Yoda is also overrated due to his particpation in tournaments like the Zotac cup. It's likely a consequence of how well korean terrans performed untill 2012, which increased their brand values, and thus they became more attractive to foreign teams as a result. Since Aliguac already is aware of the issue, I think they should have been more active in explaining the weakness of relying on certain metrics on their site, which doesn't really say alot about balance, but can confuse a lot of people.
Thank you for explaining this for me.
|
Northern Ireland23777 Posts
Outside of that what's the actual cause/ trends in the game (either facilitated by the last patch or not) that are seeing Terrans make a resurgence of sorts?
TvP it seems to have swung about a bit because outside of allins Protoss seem to be forced into opening Collosus and Terrans are adjusting their midgames to reflect this.
|
On September 02 2014 21:14 Nebuchad wrote:Show nested quote +On September 02 2014 18:03 Hider wrote:That sounds like something we could easily prove or disprove. Given how many games we add from early rounds of tournaments or qualifiers nowadays, I tend to think you're wrong there, but I can't be sure. I always wish we could add more. It's ofcourse true that the effect you will see if one race is imbalanced is that it typically will have more players going further in tournaments, and thus more games played. That effect is already taking into account by win/rates. But the effect of the average T player that participates in a tournaments being more skilled than the average P/Z-player is not taken into account in Aliguac numbers, thus some type of adjustments here needs to done. But the adjustment you're making seems too broad, given that it leads you to conclude there was TvP imbalance at a time where neither the winrates nor the results of the top players support this idea. If your model could include different performance results at different levels of skill, that would be perfect I guess, but that sounds awfully hard to do. The alternative is to be very precise in the commentary of the graph I suppose.
So I decided to just look at GSL/WCS data anyway (code S/code A + preliminearies). When looking at GSl data all you need to look at is win/rates. This is due to the fact that win/rates should increase along with the increase in player representation as the best player of the OP race will be favored against the best player of the UP race. This differs from Aliguac data which includes a lot of lower-prestice tournaments which the most succesful players won't participate in.
In the below graph the performance difference between adjusted aliguac win/rates and GSL win/rates can be seen. A positive value means Koreans are performing better, and a negative value means foreigners are performing better.
What you see is that protoss were doing fine vs terran in 2012, but not in 2011. At the end of 2012, protoss player actually had a better win/rates in GSL vs terran than according to adjusted aliguac win/rates.
Overall, however, sample size is a bit small in 2012, so its tough to say anything clearly. But from 2012-2014, you do not actually see any skill-cap differences for terran.
Protoss players on the other hand have performed better vs zergs in Korea than in the foreign scene since 2012. This kinda makes sense we actually see very few succesful foreign protosses, so perhaps protoss is the hardest race...
![[image loading]](http://i58.tinypic.com/2cyp76d.png)
Given that there was no increase in win/rates in TvZ after HOTS release, it's to me a clear indication that the quality of the data from GSL is really low, and thus there is imo very little reason to give it any more attention from a quantiative POV.
|
Even though the sample size is small, you shouldn't just dismiss it in that way.
The breaking point is when you're trying to do the adjustment to the winrates. You are saying that because there are less terrans, they play against weaker protosses on average, and so they are expected to have better than 50% results. But that isn't necessarily true. For example, the number of protosses in GSL was decreasing at this time, not the number of terrans, so there's no reason to think the people playing in GSL played against weaker protosses on average, and so there's no reason to think they were expected to have better than 50% results. And yet they did.
You are right for the recent situation, because terran was underperforming as a whole and also at the highest level, so obviously a 50% winrate wouldn't take into account that terrans played against weaker opponents as a whole. But when you do this adjustment, you have to verify that terrans did indeed play against a weaker opposition, because that isn't a necessary condition of having less players.
|
Even though the sample size is small, you shouldn't just dismiss it in that way.
I am dismissing it because it's absolutely useless from a quantitative perspective. Shouldn't we expect an increase in win/rates in TvZ after HOTS came out? Aliguac W/R clearly shows that, but WCS Korea win/rates doesn't.
It doesn't help to explain any "interesting" balance issue that Aliguac win/rates cannot answer after 2012, and since then, the sample size has only declined.
You are saying that because there are less terrans, they play against weaker protosses on average, and so they are expected to have better than 50% results. But that isn't necessarily true. For example, the number of protosses in GSL was decreasing at this time, not the number of terrans, so there's no reason to think the people playing in GSL played against weaker protosses on average, and so there's no reason to think they were expected to have better than 50% results. .
Not sure which time period you are talking about here? But in 2012 the amount of protosses increased. First season of GSL 2012 = 15 protosses in ro48. First season of GSL 2013 = 20 protosses. in Ro48.
The exact amount of players in the GSL really isn't a good indicator of balance unless it's a ridicilous low amount (like terran in 2014).
The "when you play against better players"-rule applies when the below requirements are met: (succes = actual skills + benefiting from racial imbalances)
(1) A minimum level of "succes" is required to enter the tourament
(2) Once you get above a maximum level of "succes", you no longer participate in the tournament
(3) A solid sample size
The two latter conditions aren't met for GSL, and thus it's just representation is generally a poor way of measuring balance for GSL. Instead naked win/rates are better.
|
On September 03 2014 01:42 Hider wrote:
I am dismissing it because it's absolutely useless from a quantitative perspective. Shouldn't we expect an increase in win/rates in TvZ after HOTS came out? Aliguac W/R clearly shows that, but WCS Korea win/rates doesn't.
It doesn't help to explain any "interesting" balance issue that Aliguac win/rates cannot answer after 2012, and since then, the sample size has only declined.
The problem with dismissing it is that it can lead to a situation where the best terrans win every single time, and your model says things are totally fine because the overall winrates are where you expect them to be given the population of players. That's a game Happy would want to watch, but I'm not a fan personnally.
No they weren't. First season of GSL 2012 = 15 protosses in ro48. First season of GSL 2013 = 20 protosses. in Ro48.
Oh come on, that increase is because of the people dropping from code S. Yeah, when people drop from code S, they end up in code A, that's kind of how it works =p.
|
The problem with dismissing it is that it can lead to a situation where the best terrans win every single time, and your model says things are totally fine because the overall winrates are where you expect them to be given the population of players. That's a game Happy would want to watch, but I'm not a fan personnally.
Well if it did really explain something interesting that Aliguac couldn't, then the low sample sice could perhaps be ignored, but it really hasn't explained anything since 2012. Terran haven't performed better in Korea over the last 2 years (actually they performed worse - but that's so easily attributeable to variance). I simply just don't see any advantages to continue assessing the low quality of GSL since Aliguac seems superior in every single way.
Oh come on, that increase is because of the people dropping from code S. Yeah, when people drop from code S, they end up in code A, that's kind of how it works =p.
That's reflected in win/rates then. I am not sure what point your trying to make here?
Remember I am using naked win/rates for GSL, and adjusted win/rates for Aligulac.
|
On September 03 2014 01:55 Hider wrote:Show nested quote +The problem with dismissing it is that it can lead to a situation where the best terrans win every single time, and your model says things are totally fine because the overall winrates are where you expect them to be given the population of players. That's a game Happy would want to watch, but I'm not a fan personnally.
I am only dismissing it here becasue it doens't explain anything since 2012. Terran haven't performed better in Korea over the last 2 years. Show nested quote +
Oh come on, that increase is because of the people dropping from code S. Yeah, when people drop from code S, they end up in code A, that's kind of how it works =p.
That's reflected in win/rates then. I am not sure what point your trying to make here?
Not necessarily. Let's say the population of terran and protoss are so that in your model, terran should win 65% of the time. Now let's say the best terrans win TvP 80% of the time and the average competitive terrans win 50% of the time. If you don't take into account what the best players do because of small sample size, you can end up in a situation where your model says everything is fine because TvP is at 65%, when it obviously isn't. That is a simplistic example, of course, but hopefully it helps you understand my point. You can only apply this rule of "they are playing against a weaker average opposition" if the trend of the best players' performance is the same as the trend of the average players' performance. If it isn't, then you can't just assume they played a weaker opposition as a whole, and so you can't simply apply the adjustment.
|
Not necessarily. Let's say the population of terran and protoss are so that in your model, terran should win 65% of the time. Now let's say the best terrans win TvP 80% of the time and the average competitive terrans win 50% of the time. If you don't take into account what the best players do because of small sample size, you can end up in a situation where your model says everything is fine because TvP is at 65%, when it obviously isn't. That is a simplistic example, of course, but hopefully it helps you understand my point.
Yes I understand your point, but my counter-point is that korean win/rates haven't deviated in over 2 years from Aliguac adjusted win/rates (except PvZ). Thus, it seems unlikely now that there is a noticeable difference in win/rates between top 50 players and top1000 players. That was the case in 2011, but not since 2012.
And I think its far more likely that if include GSL results as part of your overall balance view, then your going to make misassesment due to variance if for instance win/rates in GSL is 10% higher than they are in Aliguac.
So to sum up: Based on my analysis of the data I believe Aliguac win/rates today are a better indicator of highest level balance than GSL data is.
|
On September 03 2014 02:11 Hider wrote:Show nested quote +Not necessarily. Let's say the population of terran and protoss are so that in your model, terran should win 65% of the time. Now let's say the best terrans win TvP 80% of the time and the average competitive terrans win 50% of the time. If you don't take into account what the best players do because of small sample size, you can end up in a situation where your model says everything is fine because TvP is at 65%, when it obviously isn't. That is a simplistic example, of course, but hopefully it helps you understand my point. Yes I udnerstand your point, but my counter-point is that korean win/rates haven't deviated in over 2 years. Thus, it seems unlikely now that there is a noticeable difference in win/rates between top 50 players and top1000 players. That was the case in 2011, but not since 2012. And I think its far more likely that if include GSL results as part of your overall balance view, then your going to make misassesment due to variance if for instance win/rates in GSL is 10% higher than they are in Aliguac. So to sum up: Based on my analysis of the data I believe Aliguac win/rates today are a better indicator of highest level balance than GSL data is.
Sure I agree with that, because the way people have performed has been mostly uniform lately. When terran was doing badly, it was bad in korea and everywhere else. When protoss was doing well, they were doing well in Korea and everywhere else. That sure makes it easier, but it's not a given, it could change again in the future.
|
Last point I would like to make after analyzing the data: Queen patch made no sense at all balance-wise. Zergs were doing fine in terms of win/rates in both GSL (which had decent sample size late 2011/early 2012) and in terms of Aliguac win/rates.
While it made sense in terms of gameplay as early game was T-favored, it never made sense as a balance-patch and teran should have been compensated late-game while the patch was implemented.
PvZ, FYI was also Z-favored.
The patch-zerg era we saw as a cosnequence was therefore not unexpected.
Sure I agree with that, because the way people have performed has been mostly uniform lately. When terran was doing badly, it was bad in korea and everywhere else. When protoss was doing well, they were doing well in Korea and everywhere else. That sure makes it easier, but it's not a given, it could change again in the future.
Yeh I will review the data occationally, but while I orginally inteded to incorporate it into the model, I think I am gonna skip that now.
|
What do people think about the state of balance now? (the thread has been focused on historic balance for a while)
I think the widow mine buff / map change / time warp reduction / msc vision nerf have significantly reduced the number of viable openings for Protoss vs T. Seems like every Terran just goes blind Turrets and goes about their business nowadays. That addresses a huge concern that Terran players had about Protoss openings being too unpredictable and hard to scout, but I'm wondering if it's too much. Seeing way too many people straight up die because they didn't go Colossus. Not so much a balance issue as a gameplay issue though.
Especially after IEM Toronto the game looks pretty balanced, though TvP looks like it could be skewing a bit... let's see how GSL group of death looks tonight~
|
I'd consider blind Turrets bad, very bad.
It's 225 minerals for an Ebay and a single Turret, not to mention mining time.
On top of that, it does not to aid against a Blink play, I'd even say it puts you behind against one.
The threat of Oracles is still there, forcing either a Turret, 6 Marines or a Widow to be in the mineral line at 5:05 at the latest or risk losing then and there to a proxy Oracle play.
Only scouting a fast expansion or no double gas can really rule this out.
At the same time, I don't see much of any way for Terran to put real pressure on a Protoss other than forcing out a Photon Overcharge. A few Protosses do make the mistake of having little to no units at the natural (relying solely on PO, which can kill Bio but not protect Probes) and getting punished for it (last game of Flash vs Zest is an example of this I think).
Protoss can still play quite greedy and get a lot of tech out whilst staying behind PO and a few units.
Because any Terran build HAS to incorporate answers to the holy trifecta of Protoss all-ins/proxies (Oracle, Blink, DT), being aggressive is still very tricky.
As for having to go Colossus, I think this is more due to Blizzard choosing to buff existing builds via the Widow Mine instead of looking at other possibilities. In theory the shield damage buff could be removed from the Widow Mine splash but leave it intact against the primary target (otherwise Oracles / Zealots can survive single hits, making it highly supply inefficient). I'd rather see changes brought in that open up a few more options for Terran, and then modifying the Widow Mine so Templar openings have more viability.
As it is, Oracles + MSC pretty much force early Bio and nothing else.
Banshee openings are dead (Oracle detect + PO or Phoenix), any kind of Thor play died with the 250mm cannon removal and PO and any early 2rax or 1/1/1 play just dies to PO itself. Reaper plays die to enough Stalkers or the MSC.
I've seen a few Hellbat attempts but Stalkers can just kite those all day long (and they can't shoot up to help with Oracles).
If either Oracles or MSC or PO were changed, other things could be improved (or the Widow Mine could get a shield dmg nerf) but I think it would affect too many things.
Still, personally, if there were things to change I'd give these a try on a test map (not all at once):
Remove the shield damage on the Widow Mine splash, but reduce the supply cost of the mine to 1 and reduce the health somewhat (to say 60).
As it is, Widow Mines take up 2 supply. This encourages Terrans to only use it offensively because having so much supply burrowed at home weakens the army too much. At the same time, mass hordes of Widow Mines isn't a good thought either.
So, reduce the supply count but also the health (weakening offensive burrowing) and get rid of the splash shield damage. This doesn't affect Widow Mines hitting things like Oracles or Warp Prisms, but it does mean you need two Widow Mines to splash-kill Sentries or HT.
Modify MSC speed to very slow (around old Overlord speed, say 0.88) and add a MSC Speed upgrade at the Cybernetics Core for 50/50/140. This forces the MSC to be much more defensive and less out on the map.
By putting the upgrade at the Cybernetics Core, it has to compete with Warp Gate. You can either get Warp Gate or you can get MSC Speed but not both. This makes it far more risky to do a MSC poke and also puts it more at risk during a 7g Blink.
In return, the MSC with the upgrade can be faster than it currently is, giving it more mobility in the later game.
Modify the Oracle's detection ability to be at a zone, rather than on the unit.
As it is, any kind of Banshee play is completely dead because the threat of an Oracle is too great.
It can arrive more than a minute before a Banshee can, and then Protoss can just flat out make a Phoenix and kill it, Cloak or not. Protoss already has strong detection in the form of the Observer (which a Banshee can't touch) and the Photon Cannon (which can shoot both ground and air).
Furthermore, Hallucination is now free, which means a single Sentry can generate a fake Phoenix to scout everything to allow Protoss to respond in time.
However, because the Oracle denies Cloak Banshee play, Protoss can get away with forgoing detection for a long time without much risk. In WoL, going for a Blink play or 4gate without detection was risky if Terran was to go for Cloak.
At the same time, going for a Banshee meant a lot less Bio, a later expansion and later upgrades.
By changing the Oracle's detection to a zone (similar to a scan), a Banshee can still escape and live to fight another day.
The zone can protect the Probes, preventing massive Probe losses, but I do feel that if a Protoss gets caught off-guard by a Cloaked Banshee, that it should do a lot of damage, just like DTs do to an unprepared Terran.
The Banshee somewhat forces Protoss to stay at home to defend until Observers are out, as well as making it risky to move out without an Observer in the army.
Templar can feedback the Banshee instantly after being revealed (by either Observer or Oracle zone), so it might encourage some Templar play. (assuming the Widow Mine gets changed ofc)
As for any Thor/Tank play, this would require a change to the Immortal. TheDwf had a good suggestion that anything above 20 damage gets cut in half rather than reduced to 10, which Thors (30dmg) and Tanks (50dmg in siege) both do.
|
On September 03 2014 03:17 DinoMight wrote:
What do people think about the state of balance now? (the thread has been focused on historic balance for a while)
I think the widow mine buff / map change / time warp reduction / msc vision nerf have significantly reduced the number of viable openings for Protoss vs T. Seems like every Terran just goes blind Turrets and goes about their business nowadays. That addresses a huge concern that Terran players had about Protoss openings being too unpredictable and hard to scout, but I'm wondering if it's too much. Seeing way too many people straight up die because they didn't go Colossus. Not so much a balance issue as a gameplay issue though.
Especially after IEM Toronto the game looks pretty balanced, though TvP looks like it could be skewing a bit... let's see how GSL group of death looks tonight~
My opinions:
- Statistically = pretty balanced.
- Terran has too many early game options vs Zerg and Terran. Vs Zerg I would like to see a small reduction in baneling research time/research-cost. Vs zerg. it's gonna be interesting if the Snute-style becoems standard. It looks really strong so it could swing the game in the favour of zerg over the next couple of months.
- TvP is still pretty !@#$%^&* lame to watch. I really liked MC vs Flash though due to Chargelots openings being much more interesting to watch.But let's see how this matchup develops balance-wise.
- Still wouldn't mind turrets not requiering ebay. Would make it easier to be safe in TvT and TvP.
|
On September 02 2014 10:19 Hider wrote:Going into HOTS, there should obviously have been a focus on increasing the skill-cap of protoss and Zerg in order to improve overall balance.
This was painfully obvious to everyone. Even the clowns at Blizzard managed to work it out.
Just look at the new units. Both Protoss and Zerg got spellcasters (Protoss even got two if you count the mcore) while Terran got the warhound.
For better or worse, Bliz realized that introducing the ultimate a-move unit was probably not a good way to balance the game.
I will say they got the widow mine right, and they seem to have realized that as well. Requiring that Zerg splits their units in an engagement adds symmetry to the control requirements in TvZ and made that matchup much healthier. Their focus on the widow mine (especially in TvP) is their acknowledgement that a similar equalizer is required in that matchup as well - although their only accomplishment so far has been the eradication of templar openings.
Both ironically and unfortunately, the mcore probably ended up lowering the Protoss skillcap. That's a discussion for another day though.
|
|
|
|
|
|
EPT
