EPTHoping to hear from some other stats fanatics confirm or deny these claims.
Cool graphs tho bro. =]
Statistical Analysis of StarCraft 2 Balance - Page 2
| Forum Index > SC2 General |
|
MusiK
United States302 Posts
Hoping to hear from some other stats fanatics confirm or deny these claims. Cool graphs tho bro. =] | ||
|
duk3
United States807 Posts
Interesting read. How did you determine the B of each player? | ||
|
Mordiford
4448 Posts
| ||
|
wherebugsgo
Japan10647 Posts
You might want to go back and fix these things. (I actually haven't taken the time to fully understand the statistics, so I apologize in advance for not providing further insight other than the presentation of the paper) example: in Appendix A the italicized names of Protoss and Terran are spelled wrong. | ||
|
Drowsy
United States4876 Posts
On May 05 2011 09:54 Malpractice.248 wrote: Wait, you think Z > P, at all levels, or on average? At top lvl, its quite a diff story... The sample is 852 GSL games, but it goes back to October and obviously there were patches. I found it a pretty good read and it points to the game being relatively well balanced. | ||
|
Techno
1900 Posts
On May 05 2011 09:41 d_ijk_stra wrote: The difference of this article and previous one is that here I tried to take each gamer's ability into account. That is, oGsMC dominating Terran gamers does not necessarily mean that P >> T, since it could be that actually TvP balance is quite good but MC is just too strong. By taking each gamer's ability into account (using what statisticians call 'latent variable'), I think I resolved this problem. Uhhh unfortunetly this sounds brutal. You cannot define skill. Period. You should have assumed it's normally distributed among the population. | ||
|
Al Bundy
7257 Posts
Is it explained somewhere? I may have missed it | ||
|
d_ijk_stra
United States36 Posts
I think you confused it with random effects / hierarchical model in ANOVA. You don't really need to assume latent variable to follow normal distribution. Of course, without any regularization it will overfit data, and using the assumption of normal distribution is a good way to regularize your parameters. But you can also use other types of regularization... I used L1 penalty for other reasons. However, I guess you may not want to discuss this much of technical details  | ||
|
d_ijk_stra
United States36 Posts
| ||
|
Primadog
United States4411 Posts
| ||
|
Saechiis
Netherlands4989 Posts
I'd definitely add a note that it's an approximation of racial balance as seen in GSL games played up until March. It doesn't say anything about racial balance now, so you can save this thread a whole lot of "I told you X is OP and Y UP, this proves it!". | ||
|
Carkis
Canada302 Posts
| ||
|
Quochobao
United States350 Posts
But to everyone else, the statistical significance is LOW, meaning that there is too much variance, and the result may be MERE CHANCE. Please don't say anything about balance based on this >.< | ||
|
Nontrivial
United States56 Posts
Here is the link to what I'm referring to: Link | ||
|
TheRabidDeer
United States3806 Posts
On May 05 2011 13:44 Techno wrote: Uhhh unfortunetly this sounds brutal. You cannot define skill. Period. You should have assumed it's normally distributed among the population. You can define skill, and when working with a definite set of games with a limited number of players you probably should define skill. He is working within the confines of probably 50-100 players, since he is using GSL games. While that is "statistically" a large enough sample size, it could easily produce errors when there are so many other variables included. Example: some players suck at a certain map or matchup which skews the data. This brings up a question to the OP though: I know you calculated player skill and map balance, but did you attribute a certain players matchup skill and their ability for a certain map? Or is there not enough data or is that too complex a situation to work out the numbers? ie: MC may have a very strong PvT on xel'naga but a very weak PvT on tal'darim (dont know if its true, just an example). A different map may require different strategies in a certain matchup that the player isnt necessarily as good with. | ||
|
Zedders
Canada450 Posts
It'd be interesting to see what the average game length is over time as well. Since cheeses have changed a lot since the game started (5rax reaper and whatnot), people have a) learned to deal with cheeses all-ins more adequately and b) developed more late game strategies, the games are probably as a result, longer. It isn't surprising to see that terran was so dominant at the beginning because of the number of people that started out playing terran. If i recall...the first GSL was vastly Terran populated. Not to mention vastly cheese populated too  Terran of course having the strongest tier one unit, the marine, had (has? i'm not sure anymore) the strongest early game. We all of course remember the BitByBit strategy (essentially all-inning...and if that all in doesnt work...all in again....and if that doesnt work...all in again...rinse and repeat). Since terran had the strongest early game...the game ended fast because cheeses were so powerful/prevalent. Therefore Terran won a lot. The games are getting longer now.... this of course results in more and more mistakes made by each player. Balance, in my opinion, should be weighted on how many mistakes the player can make in proportion to the other player's mistakes. What I mean by this is if one player makes less mistakes in his game decisions, he should ultimately win in a long game. Why you ask? Because Starcraft 2 is a game of decisions. And the longer the game goes on, the more decisions must be made. The more decisions that are made, the more mistakes there are, which should result in the degree of separation that makes one player better than the other. In context...let's say X race gets supply blocked 2 times (common macro mistake) but Y race never gets supply blocked. Y race then as a result has a larger army, larger economy etc. X race still wins simply because the units he made counter the units Y race made. Ok...this isn't imbalance...this is strategy right? Y makes a larger mistake by not scouting X and as a result his units crumble to X's. So we've established that theres different TYPEs of mistakes one can make. And some mistakes are weighed less than others. But at what point do these mistakes balance. What if X can get supply blocked twice, not scout opponent's army (+more mistakes) and still win. The severity of one race's total mistakes should not be much larger than another's. Ultimately I'd like to see X -not- win and I hope you agree with me, because X is clearly not the better player, his race is. --------------back to the graphs..... Ok so these graphs are representations of both races making an equal amount of mistakes since they are pros, and we are assuming that most pros compete at the same skill level regardless of race. So the degree of seperation of skill because of the mistakes that are made should be negligible. To sum up a little....... The average game length has increased (I'm pretty sure of this considering map size, cheese prevalance, spawn points). More game length means more potential for mistakes. Ultimately as e-sports fans, we want to see the better player win. This means the player that made the right call at the right time, with the right micro, while maintaining the right macro. Now it's super important to note...these graphs don't display anything about HOW the games were won. Looking at T v P... you might think "oh look it's balanced now because it's 50%/50% wins now" November2010 to jan 2011....Terran cheese prevails until protoss finally learns how to stop it (or they patched whatever). The game was balanced in january 2011 because Protoss learned how to stop strong terran all-ins? (the emergence of a 'safe build' to gain eco lead was developed) this isn't balance, this is metagame development, meaning half the people that are trying the old strategies that used to work 60% of the time, failed a lot. And the other half that realized this, tried new strategies (and not as developed and therefore not as good) won because it was something their opponent hadn't seen before. yay for meta game development! | ||
|
huameng
United States1133 Posts
Thanks for doing this by the way, it's really interesting. | ||
|
d_ijk_stra
United States36 Posts
TheRabidDear/ This is a good point, and I strongly agree that gamer-map interaction should be taken into account. It really needs to be done, but as an initial investigation I wanted to keep the model simple and stopped at this point. Well I have a lot more ideas to make this model much more realistic, but as a graduate student I need to work on the project which pays me... I did it during the spring break If there's adequate conference/journal then maybe I can spend more time on it... If anyone has some idea, please let me know. | ||
|
arbitrageur
Australia1202 Posts
"It is Xel'Naga Caverns, and it turns out that the map favors Zerg slightly over Terran" Says who? Says you? You realised ladder data cannot be extrapolated to the highest levels because the vast majority of people on ladder do not know how to play. | ||
|
d_ijk_stra
United States36 Posts
On May 05 2011 14:35 huameng wrote: How did you estimate the Beta values? Sorry if it's in the paper and I missed it, but I didn't see anything. Thanks for doing this by the way, it's really interesting. huameng/ you may refer to my comment to Albundy | ||
| ||
