|
On January 11 2014 02:30 Hider wrote:Show nested quote +
Zerg had virtually no aggressive (read: early-game) options vs Terran in Broodwar but everyone just accepted it because that's what the game was.
I actually think the reason for why terrans complaint isn't that they can't put reliable pressure pre 10th minute mark on the protss. But rather, that protoss has so many insanely powerfull allins and terran has 0. So its the combination really. Now, if protoss allins were "just" as strong as in WOL, the assymetry would be lower and people wouldn't complain as much. In TvZ BW - Terran didn't really have a wide variety of pre 10th minute allins that were super hard to beat, and that's the difference.
It's pretty much the same situation that existed in TvZ in early WoL. Zergs were angry because they felt, not unjustifiably, that terran could just make more or less whatever and attack, whereas zerg has almost no aggressive options outside of complete all ins. It's one of those places to be for a game, where even if it's balanced it's still not good, like Vortex vs Brood Lords.
Also early game TvZ in BW was very mapped out, zerg was rarely in any real danger unless it was a cheese, and it was understood that terran needed to put some pressure on not to fall behind. I still think Irradiate was bullshit though :p
|
|
|
China6332 Posts
Also I should point out that Team 1 the RTS team is also making Heroes of the Storm. I think given the market trend the majority of the development resources has been assigned to Heroes instead of Starcraft, which explains the unusually slow reaction to feedback, such as the league distribution/MMR decay issue, because they simply don't have the resources to make bigger changes.
|
On January 11 2014 03:02 digmouse wrote:
Also I should point out that Team 1 the RTS team is also making Heroes of the Storm. I think given the market trend the majority of the development resources has been assigned to Heroes instead of Starcraft, which explains the somewhat unusually slow reaction to feedback, such as the league distribution/MMR decay issue, because they simply don't have the resources to make bigger changes.
A nice way of saying they hardly give a shit anymore and are moving on to the next thing. Hopefully they don't forget that Legacy of the Void stands to reinvent the metagame and isn't by any means less important than some DOTA clone.
|
United States7483 Posts
On January 10 2014 11:47 seak99 wrote:
what do you think he means by this :
"Please keep in mind these are not straight-up win percentages. They’re win percentages with player skill factored out. When we grab win/loss data for balance purposes, we categorize each game with 2 different variables per side: one being player skill and other being race strength. So by factoring the player skill out, we are able to more accurately check how each race is doing at each skill level."
It means they run econometric analysis of their data in order to determine how much of the results are due to which variable, then factor out the player skill variable to see where balance lies. It's mathematically possible.
|
It means they run econometric analysis of their data in order to determine how much of the results are due to which variable, then factor out the player skill variable to see where balance lies. It's mathematically possible.
What variables are they going to use to forecast skill level? EAPM? SQ?
If they do, then tons of issues arive which I am not sure you can fully account for. For instance imagine this:
Prediceted win rate= EAPM *X + SQ * Y + racedummy*EAPM + racedummy*SQ + playerspecific MMR * Z.
Now here is the usse. How did they arrive at hte values for X and Y, and the values for the racedummies?
If they are basing this on current data, then the data's are already biased since they already can sufffer from balance issues. E.g. if protoss is imba today, then protoss will get a higher predicted winrates through its "racedummy".
In order to get reliable data, they need to use a period of time where they know the game is balanced. Since we don't have that (we can never really know), then econometric analaysis is absolutely nonsense here.
Its just more more reaslitic to make adjustments based on distribution instead of this.
|
United States7483 Posts
On January 11 2014 03:19 Hider wrote:Show nested quote +It means they run econometric analysis of their data in order to determine how much of the results are due to which variable, then factor out the player skill variable to see where balance lies. It's mathematically possible. What variables are they going to use to forecast skill level? EAPM? SQ? If they do, then tons of issues arive which I am not sure you can fully account for. For instance imagine this: Prediceted win rate= EAPM *X + SQ * Y + racedummy*EAPM + racedummy*SQ + playerspecific MMR * Z. Now here is the usse. How did they arrive at hte values for X and Y, and the values for the racedummies? If they are basing this on current data, then the data's are already biased since they already can sufffer from balance issues. E.g. if protoss is imba today, then protoss will get a higher predicted winrates through its "racedummy". In order to get reliable data, they need to use a period of time where they know the game is balanced. Since we don't have that (we can never really know), then econometric analaysis is absolutely nonsense here. Its just more more reaslitic to make adjustments based on distribution instead of this.
I don't know what data they are using, what process they are using, etc. Making any judgments at all without access to their methodology is impossible, and there are ways around the problems you stated. Since they are not a science group publishing an article, we're not going to get their data. We can either assume they are all incompetent and know jack shit, which is what most people seem to want to think, or that maybe they have some actual reasons to behave the way they have been doing, and that we might not actually know as much as we think we do.
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
|
There is no way on make a math equation to detect and measure skill. You might as will try to build one to detect if someone is a good football or hockey player based on film and player stats. People try and all the systems they create are far from accurate.
|
On January 11 2014 03:30 Plansix wrote:
There is no way on make a math equation to detect and measure skill. You might as will try to build one to detect if someone is a good football or hockey player based on film and player stats. People try and all the systems they create are far from accurate.
Let me ask you a question... is the Easy AI better (in terms of being able to defeat X opponent) than the Hard AI in SC2? I imagine the Hard AI would win, most, if not all the time against the Easy AI.
There we have quantified skill, perhaps 100% accurately. We can certainly do that with players too. Their accuracy however, is subject to many more variables. If we could control those variables, or at least account for them, we certainly could create a system that was 100% accurate. Doing so is likely impossible, but we can achieve a decent rate of accuracy through mathematical systems. And this has been shown.
Remember, Vegas would never allow people to bet on sports games if they didn't have some degree of certainly that one side was going to win.
I do think, however, that qualitative methods are actually what SC2 is in dire need of. Not just in terms balance, but game design. Sure, XvY matchup might show a 50-50 win rate, and appear balanced. But if the game is boring and there is no variety, then there are problems with game design that qualitative methods could find and then these issues could be quantified. For instance if every PvP is 4 Gate vs 4 Gate qualitative methods would discover this and to the degree that 4 Gate vs 4 Gate happens. This takes a lot more work than quantitative analysis and depends on the skill of the researcher, but quantitative methods alone can not discover balance problems stemming from game design (such as X unit being underpowered and never used because Y unit is too strong).
I did a significant amount of qualitative research when I created this thread: http://www.teamliquid.net/forum/viewmessage.php?topic_id=378373
The results speak for themselves. Months before the changes, I predicted Hellbats were OP (comparable to BFH), Tanks needed their attack speed increased, ect, ect...
|
United States7483 Posts
On January 11 2014 03:30 Plansix wrote:
There is no way on make a math equation to detect and measure skill. You might as will try to build one to detect if someone is a good football or hockey player based on film and player stats. People try and all the systems they create are far from accurate.
You make estimates, evaluate through tests how far off from your predictions the results are and determine whether they fall within your margins of error or not. If they do, given a large enough sample size, you can reliably conclude that the negative is untrue (rejecting the null). So yeah, you actually can, sort of. You're not trying to determine the skill of a single player, you're trying to determine the average skill level of the zerg pros vs. terran pros. vs protoss pros on a large level, and you can get reasonably close. You also don't need an actual answer, just need to know how to factor it out.
The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now.
EDIT: Bronzeknee probably did a better layman explanation than I did.
|
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation.
But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race.
The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now.
No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed.
|
United States7483 Posts
On January 11 2014 03:47 Hider wrote:Show nested quote +
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation. But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race. Show nested quote +The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now. No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed.
I'm telling you that you cannot possibly know what variables they do and don't have, or even what datasets they do have. Clearly they seem to think they know more about it than you do, and I'm inclined to believe that they probably do.
|
Maybe off topic but, is there still a bug with the ladder(s)? I see people posting negative bonus pools? What's up with that?
|
|
|
On January 11 2014 03:55 Whitewing wrote:Show nested quote +On January 11 2014 03:47 Hider wrote:
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation. But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race. The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now. No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed. I'm telling you that you cannot possibly know what variables they do and don't have, or even what datasets they do have. Clearly they seem to think they know more about it than you do, and I'm inclined to believe that they probably do.
Well can think the exact same way that they do. So given that we worked at Blizzard, how would we proceed about making an ecometrical model.... And I am telling you that if you worked at blizzard and got reported the ladder statistics, they would be biased. Now, how are you going to clean it? How would you proceed, and then next which explanatory variables are you gonna use.
Unless someone can come up with a great answer for that, it just doesn't make sense at all to believe that there is some fantastic econometric solutions that gives us a great sense of Blizzard.
|
On January 11 2014 03:30 Plansix wrote:
There is no way on make a math equation to detect and measure skill. You might as will try to build one to detect if someone is a good football or hockey player based on film and player stats. People try and all the systems they create are far from accurate.
I don't believe it is possible either, not in any substantive way (especially when applied to as complex an organism as Starcraft). But, once the game is an E-Sport and once there is a balance team in operation, a metric has to be devised to guide the team's actions. 50/50 is one way after which comes a method to best decide 50/50 and how it is applied.
The method would have been devised and agreed and signed off by concerned departments and senior management. The scientific soundness of the method is irrelevant. Whether it is coherent and agreed internally is the issue. And it is.
|
United States7483 Posts
On January 11 2014 04:11 Hider wrote:Show nested quote +On January 11 2014 03:55 Whitewing wrote:On January 11 2014 03:47 Hider wrote:
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation. But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race. The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now. No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed. I'm telling you that you cannot possibly know what variables they do and don't have, or even what datasets they do have. Clearly they seem to think they know more about it than you do, and I'm inclined to believe that they probably do. Well can think the exact same way that they do. So given that we worked at Blizzard, how would we proceed about making an ecometrical model.... And I am telling you that if you worked at blizzard and got reported the ladder statistics, they would be biased. Now, how are you going to clean it? How would you proceed, and then next which explanatory variables are you gonna use. Unless someone can come up with a great answer for that, it just doesn't make sense at all to believe that there is some fantastic econometric solutions that gives us a great sense of Blizzard.
You do realize there are econometric tests for measuring bias in variables, which then allow you to adjust for it, right? You also don't know they are working with ladder statistics necessarily, maybe they're working with some kind of weird GM+high master and pro match model, or they have a different stats page for each league, or whatever.
|
On January 11 2014 03:55 Whitewing wrote:Show nested quote +On January 11 2014 03:47 Hider wrote:
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation. But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race. The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now. No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed. I'm telling you that you cannot possibly know what variables they do and don't have, or even what datasets they do have. Clearly they seem to think they know more about it than you do, and I'm inclined to believe that they probably do.
You believe the data which shows balanced win-rates at gold, silver and bronze in PvT?
I can accept that PvT is pretty balanced at the top level, but everyone knows that Protoss is clearly favoured in a-move and all-in scenarios - which makes up the majority of playstyles at lower levels.
Can you explain how bronze, silver, and gold PvT is pretty balanced? Because I am not convinced it should be balanced given what we know about Protoss deathball vs Terran bioball.
(note: I'm not arguing that PvT should be balanced at lower levels.)
|
On January 11 2014 01:45 Grumbels wrote:Stats from aligulac for the WCS Korea Code A qualifiers: PvT 32–30 (51.61%)
PvZ 47–43 (52.22%)
TvZ 35–37 (48.61%)
35 PvP, 11 TvT, 40 ZvZ It's the same pattern: terran is slightly disadvantaged in both non-mirrors and there is a general lack of terran players to begin with.
Have you seen the draw? That was just purely random as of how many mirror matches that are played. There were some groups which were almost exclusively 1 race, which results in a huge amount of mirror matches automatically.
1st Session: 23 Terrans, 22 Zerg, 26 Protoss
2nd Session: 21 Terrans, 27 Zerg, 25 Protoss
So... yes, there were slightly less Terrans in the qualifier (44 vs 49 vs 51)... but we had 2 more qualified for GSL already. And the Terrans didn't do very poorly in the qualifier considering once again the fact they had the most players pre-qualified. We have an almost even distribution in GSL for next season, so I don't think the GSL qualifier is a good basis for an argument :-P
|
United States7483 Posts
On January 11 2014 04:14 plogamer wrote:Show nested quote +On January 11 2014 03:55 Whitewing wrote:On January 11 2014 03:47 Hider wrote:
Also, why the hell would you make EAPM a variable? That's never been indicative of skill. Remember when good ol' 100 APM Sjow beat life? I would never have the equation look anything like that at all.
You were talking about econometrics. An econometric analaysis is without a doubt gonna find a significant correlation between EAPM and skill-level. I think your confusing no outliers w/ correlation. But as you say, its not perfect indicator, and that's the problem: There are no good objective measures of skills (no explanatory variables) other than looking combined at MMR and distribution of the race. The end result is that you can come to a reasonably close estimate of what the game balance actually is at a given point in time, even if it's not perfectly accurate. Their stated goal is to keep balance within the 55-45% range for winrates, they're doing much better than that right now. No you can only do that if you have some really explanatory variables and a dataset that is cleaned for balance. Blizzard simple doesn't have that (as I argued in the previous post). All of the variables they will use are simply bias'ed. I'm telling you that you cannot possibly know what variables they do and don't have, or even what datasets they do have. Clearly they seem to think they know more about it than you do, and I'm inclined to believe that they probably do. You believe the data which shows balanced win-rates at gold, silver and diamond in PvT? I can accept that PvT is pretty balanced at the top level, but everyone knows that Protoss is clearly favoured in a-move and all-in scenarios - which makes up the majority of playstyles at lower levels. Can you explain how bronze, silver, and gold PvT is pretty balanced? Because I am not convinced it should be balanced given what we know about Protoss deathball vs Terran bioball. (note: I'm not arguing that PvT should be balanced at lower levels.)
Probably because in bronze silver and gold, protoss player suck horribly at defending terran all-ins too. It's very hard to even find balance relevant at those levels where everyone is playing extremely sub-optimal. I remember watching Totalbiscuit win a bunch of games with a 1 base battlecruiser rush in gold. That's in no way indicative of whether balance exists or not.
|
|
|
|
|
|
EPT
