|
I think 'moneyballing' is impossible when you look at things like creep spread.
If you find a positive correlation between the amount of creep tumors and winrate, it still doesn't mean there is causation. A higher amount of creep tumors can also be caused by the fact that the player is winning or is in a better position. For example, when you are in a winning position you usually spent less APM on defending which leaves more free attention for active creep spread.
So, winning leads to creep spread? Or the other way around? You just don't know until you identify and test the underlying variables, which is pretty hard i guess.
Same goes for the example that said people that buy upgrades win more: when you're in a winning position you usually have more resources available for upgrades.
Obviously you can get some decent statistics from analyzing replays, like: when you build no workers, odds are you are going to lose. Which is probably true, but not very useful.
|
As said in a couple previous posts Starcraft is a reactionary game. You observe what you're opponent has done or not done then react accordingly.
A couple have said that it's possible to retain every build. While the generic early-to mid game builds can memorized quite easily the hundreds of variations in those builds is what makes it complicated. If all it came down to was builds then anyone could be a pro. Just execute your build and you win right? Wrong it's the variations in the builds that makes the pro distinguished from the non pro player.
For example micro, micro is used by even the lowest level player ranging from different difficulties, wether it be the drone split or a complex move with your army it's used by most. Now we look at the professional player and we execute the same build he used against the same build the oponent used against the pro. However in our case we lose and the pro player wins. What went wrong? We both did the same build lets even go so far as to say our macro was on the same par.
The loss on our part could of been attacking in a disadvantagous choke point, it could have been not having that extra couple of units we needed, it could of been as simple of attacking when we weren't supposed to. The fact is that to complete your statistics in way that the article OP was mentioning you would literally have to take into consideration every single variation that the opposing player can do. If he makes an extra two immortals will I die to this push? What about 1? If he makes two how much would that take from his army? What if he doesn't make any units and just turtles and I prepared for a big push that didn't come just because my oponent was nervous even though the build executed was supposed to attack at this time. In fact it's not irrational to say that sometimes if we make a mistake like droning when we should be making units that could actually become beneficial in some circumstances. If you honestly think its possible to take every single possibility into account think about a Match up and then the main starting builds that come into play. Then think about the branching possibilities that can branch out of each build and then on top of that think of every unit composition that's possible from those branching possibilities. After that think of the timings then take into consideration places that would've disadvantagous and advantageous to attack at. It goes on and on. Not to mention the different maps.
Edited for clarity 
|
Now we get to the individual sport of Starcraft 2. The author suggest creep spread as an interesting metric to look at to see if he's a good player. Let's say that we found out that zergs who average placing 25+ creep tumors per game win an average of 55% of their tournament games and ones who placed less than 25 on average only won 45% of their tournament games. So, as a team manager, you'd go grab all the data for tournament players and try to find the one that places 25+ creep tumors per game and hire them, right? WRONG! You have a better statistic called win percentage right in front of you. A player's value as a winning player can be sumarized by win% because the only other factors that effect it are map and opponent. When it comes to creep spread, you still have map and opponent as a factor, but you also add a ton of other major factors and probably an infinite number of minor factors before you get to a win/loss.
I think you are missing something here. Sabermetrics' goal was to increase win percentage. I think the author was eluding to specific aspects, such as creep spread, can be compared to other players and you might be able to find certain aspects more important than others. For example, a player who spreads 10 more creep tumors than another increases win percentage by 1%.
If you are actually able to determine parts of a players game that could be improved so that they can then win more games then that player knows where and what to practice. It's not really the same, but both sabermetrics and moneycraft are both tools with the goal of improving win percentage.
edit: a better example than creep spread that is easier to see is the average time in between spits. Usually the higher ranked you are the less time there is.
|
On February 18 2012 06:05 drgoats wrote: For example, a player who spreads 10 more creep tumors than another increases win percentage by 1%.
This is where i think alot of people are gettin misconstrued. They will take one statistic and say that, that makes a player the extra percentage to win. However think about the individual spreading creep, like someone said previouslly the creep spread could be due to the fact that he has not been pressured at all. He has the liberty to spread creep more so he does it. You cant take a single element of data and just declare this the leading factor as to helping the person win. The person who has amazing creep spread more than likely also has amazing macro and since he is so used to macroing at an easy pace he can also multitask the creep spread. There it would be the macro that would be the deciding factor and not the minor creep spread. You're trying to look at the results and conduct a reason why the person is winning when you should be looking at the roots AND the results.
APM is a HUGE one of these misconsceptions. I think that its died down more now but in the past people were freaking out about APM. APM became the deciding factor in skill among lower ranked players. The reason why higher level players did not stress over APM is because they knew that even if you APM is 300 that if the APM is not put into anything usefull then the APM really does not matter.
|
On February 18 2012 04:57 dementrio wrote:
I might be too old to understand, but is this "number crunching" what they used to call "data mining" in the olden days? You know, the diaper & beer story?
"number crunching" to me is a term so general I can't understand what you are talking about.
Number crunching includes data mining, but in this case has more to do with mathematically predicting the outcome of every action and the efficacy of every build against every other build. This information isn't hidden away somewhere to be found, hence why it isn't called data mining.
|
On February 18 2012 06:16 Doomville wrote:Show nested quote +On February 18 2012 06:05 drgoats wrote: For example, a player who spreads 10 more creep tumors than another increases win percentage by 1%. This is where i think alot of people are gettin misconstrued. They will take one statistic and say that, that makes a player the extra percentage to win. However think about the individual spreading creep, like someone said previouslly the creep spread could be due to the fact that he has not been pressured at all. He has the liberty to spread creep more so he does it. You cant take a single element of data and just declare this the leading factor as to helping the person win. The person who has amazing creep spread more than likely also has amazing macro and since he is so used to macroing at an easy pace he can also multitask the creep spread. There it would be the macro that would be the deciding factor and not the minor creep spread. You're trying to look at the results and conduct a reason why the person is winning when you should be looking at the roots AND the results. APM is a HUGE one of these misconsceptions. I think that its died down more now but in the past people were freaking out about APM. APM became the deciding factor in skill among lower ranked players. The reason why higher level players did not stress over APM is because they knew that even if you APM is 300 that if the APM is not put into anything usefull then the APM really does not matter.
I totally agree that it is a lot more complicated than the example I used, I just used it because it was in the article.
To counter your argument, the A's won 103 games that year because they had a great offense. It works the same as your macro statement. It is once you start breaking it down that you can see key factors at why they were winning, especially in the players that they chose. You can go really simple and then complicate it from there. Worker counts is a good example.
In a >20 minute game in TvZ, with the zerg on 3 bases, what is the amount of workers that the zerg player has to bring the highest win percentage.
Edit: oh, and I agree with your APM point. But that is why people are trying to make a more useful stat with EAPM or CPM (Blizzard's idea). I have seen a direct correlation between my improvement and my EAPM.
|
What happened in moneyball and what you are talking about in Starcraft 2 are seperate things.
In moneyball you are drafting a team that has the highest chance to win based on statistics. This cannot be compared to estimating or raising the chances of a player to win. I believe you could only use the same techniques in moneyball to make a team (for GSTL perhaps) which has the highest chance to win based on statistics.
Another view of this is that, in moneyball, the players do not need to care about or even know about what the statisticians are doing to construct their team. They just play as they usually would besides for general advice from the coach/statistician such as "try to get on base via a walk". It will only work in this same fashion in SC2. For example, a coach could tell one of his team's lineup to "try to win this game via 6 minute rush" for this to be in parallel with moneyball. This type of "play" could statistically cause the other team to adapt their strategy a certain way, cause the other play to tilt or just downright lose. The player's surely don't need to be aware of the statistics just as much as the statisticians do not need to play baseball/sc2.
Hope this is clear as it is a complex subject.
|
Sports statistics is more about asking the right questions rather than tools or mathematics depth. How would we go about numerically measure concepts like micro and macro?
|
On February 18 2012 06:38 willoc wrote:
What happened in moneyball and what you are talking about in Starcraft 2 are seperate things.
In moneyball you are drafting a team that has the highest chance to win based on statistics. This cannot be compared to estimating or raising the chances of a player to win. I believe you could only use the same techniques in moneyball to make a team (for GSTL perhaps) which has the highest chance to win based on statistics.
Your point is spot on and moneyball is definitely not the right term. Just saying that people should use data from other player's replays would have been better. However, with the movie coming out last year it is a lot more fun to use the comparison.
Also, if you look at it like you as the player is the team and you are using data from the player pool (drafting) to formulate your approach to SC2. For example, there is probably a certain ratio between fighting units and workers that will bring you as a player more wins than with other ratios.
You can also look at a comparison between micro and macro. Some time spent on micro is good but how much? SC2Gears breaks it down for you and then you can compare yourself to the top players. If the average for top players were spending 70% of their apm on macro and 30% on micro then I know that I should adjust my ratio to match theirs.
|
On February 18 2012 06:59 drgoats wrote:Show nested quote +On February 18 2012 06:38 willoc wrote:
What happened in moneyball and what you are talking about in Starcraft 2 are seperate things.
In moneyball you are drafting a team that has the highest chance to win based on statistics. This cannot be compared to estimating or raising the chances of a player to win. I believe you could only use the same techniques in moneyball to make a team (for GSTL perhaps) which has the highest chance to win based on statistics.
Your point is spot on and moneyball is definitely not the right term. Just saying that people should use data from other player's replays would have been better. However, with the movie coming out last year it is a lot more fun to use the comparison. Also, if you look at it like you as the player is the team and you are using data from the player pool (drafting) to formulate your approach to SC2. For example, there is probably a certain ratio between fighting units and workers that will bring you as a player more wins than with other ratios. You can also look at a comparison between micro and macro. Some time spent on micro is good but how much? SC2Gears breaks it down for you and then you can compare yourself to the top players. If the average for top players were spending 70% of their apm on macro and 30% on micro then I know that I should adjust my ratio to match theirs.
Fully agreed.
I'm wondering, maybe it would be most useful if players were first classified in some way (creative, macro, micro, harassment) based on their key strengths. Then with a grouping of those players, it could be seen what is the most beneficial for them to concentrate on to raise their' chances to win? Examples of such concentrations (with metrics) could be creep spread (# of tumors), # of expansions per minute, # of production facilities per minute, army compositions (# of or ratio of types of units), etc. These could even be further divided down by periods of the game (begining, middle, end) or tiers. I would not set minute markers to differentiate between these periods of the game but rather the tech level and amount of production facilities as thresholds. What do you think?
|
On February 18 2012 00:54 FFGenerations wrote:
valkyries man
User was warned for this post
oh im so sorry for not adding to the discussion, forgive me, hope i didnt spoil anyones night
|
On February 18 2012 04:42 Primadog wrote:
RenSC2, wouldn't the team leagues be sufficiently close to the hybrid sport you refer baseball as?
It is another form of a hybrid individual-team sport, but it goes about being a hybrid sport in a different way. It doesn't work the same as baseball because each player gets a win or loss based on his individual performance. For example, Jaedong goes out for Hwaseung Oz and either wins or loses his match. Whether the rest of the team wins or loses the match is irrelevent to how we evaluate Jaedong. We don't need to go in and measure his APM or his worker count at the X minute mark or whatever. We already have the final product available to us: Win%. Then we only need to adjust that for opponents/maps and we know how good Jaedong is. As everyone knows, he was very good despite a weaker team.
In baseball, traditional managers looked at players as winners or losers because they were on winning or losing teams in the past. Or they overvalued the wrong stats. Batting Average was the major hitting stat along with Home Runs, but they've more recently figured out that On Base Percent tends to be a more important stat. Likewise, a pitcher's win count was seen as one of the most important pitching stats. However, his wins/losses were highly effected by the actions of his teammates. Unlike a Starcraft player's Win%, a pitcher's Win% was not a result of his efforts alone. That's why you need to look deeper into a pitcher's stats to determine his real effectiveness.
And of course, I do want to repeat that statistical analysis (or whatever you want to call it) can still be important for seeing how to improve your game. But that typically means playing more like top players, rather than using the numbers to innovate. I would just reject anything like Moneyball for creating a SC2 team. It makes no sense when you have win% as such a readily available stat.
|
|
|
|
|
|
EPT
