Zerg rule finally toppled - Page 9

On April 19 2013 09:19 AndAgain wrote:
OP, why don't you use these methods to get to #1 on liquibets? It would give you more credibility.

On April 19 2013 08:45 IcedBacon wrote:
Well that list looks absoutely inaccurate considering the koreans that are on it. Life is not surprising but the rest are. I wouldnt use this data for anything, TLPD ELO is way better.

On April 19 2013 16:48 Zealously wrote:
But guys why is fOrGG #4?

On April 19 2013 17:04 ninazerg wrote:
According to your statistics, there are 5 zergs in the Korean top 10, 1 protoss, and 4 terrans. How is that "Zerg rule finally toppled"?

On April 19 2013 17:01 SlixSC wrote:
It doesn't matter that you know your system is inaccurate and somehow use that as a self-justifying argument for posting these statistics anyway. If I were to come up with a system that would show forGG as the 4th best players in the world I would be so embarrassed and definitely NOT post them anywhere on a public forum.

Statistics done poorly by people who are aware of the flaws in their system but post them anyway and make it sound like they have any descriptive power.

I don't even...

On April 19 2013 17:07 Grovbolle wrote:

Actually, in general, over a large period of time, the aligulac system have better predictive power than the average liquibet players as well as better than any system currently out there.

Our ranking is not the ultimate truth, nor has it ever been. We are always working on making it better, and one of the issues we are very aware of are the problem with "closed pools" like the ones forGG have getting his surge in points from.

However just because it has flaws, it doesn't mean that you can't still use it, and with some common sense and supplementing with your own knowledge of the scene, try to make a fair who is the best in the world opinion.

Actually aligulac should probably be seen as a site which rewards players who have been playing good in a longer/shorter period of time and are therefore rewarded with points from players who have underperformed.

As always, I encourage people with statistical knowledge to download our DB mysql dump and try out building their own models so we can get a healthy discussion and ultimately, hopefully, improve the entire system.

On April 19 2013 17:11 SlixSC wrote:


Being more statistically accurate than TLPD is like taking candy from a baby.

You have found a terrible system and made a slightly better one.

*standing ovations*

On April 19 2013 17:19 Grovbolle wrote:

Yeah that is not what I was saying at all. I simply stated that while our system is far from perfect, it does have a higher predictive power than eg. Liquibet players, and that is without taking stuff like map balance in to consideration.

I am glad that you are able to rank players with your own perfect knowledge of their skill level. We simply try to do based on whatever results are available to us from various tournaments while also building a system which is also capable of predicting outcomes of games.

As always, if you are capable of making a better system, we encourage you to do so, we are already doing all the hard work, which is collecting and organizing the matches as well as categorizing stuff like online/offline and WoL/HotS and making it available to the public so they themselves can experiment with it. If not, then please give suggestions as to what can be improved regarding the ranking method and how you propose we do that.

On April 19 2013 17:35 SlixSC wrote:


The fact that nobody has come up with a better system doesn't justify the fact that your system is inaccurate.

You seem like the kind of person that would say things like.. "Yes I failed the test, but so did my best friend and he had even less points so it's all good!" Except for the fact that you failed of course.

How predictive is the system????

By my understanding, very.



This is a plot of more than 100k historical games (the whole database as of February 2013). On the horizontal axis you find predicted winrate for the presumed stronger player, using the ratings at the time the game was played. The games were grouped in reasonably small groups, i.e. 50%-53.3% and so on upwards. (Obviously no numbers below 50% since this is the predicted winrate for the stronger player.)

On the vertical axis is the actual winrate for each group.

As can be plainly seen, the actual winrate is close to the predicted winrate up to about 80%. The weighted linear fit (blue dashed line) is almost identical to the ideal curve (red dashed line) corresponding to the relationship that predicted winrate equals actual winrate.

Above 80% the winrates become a bit more erratic, and it seems like the system slightly overestimates the chances for the stronger player.

Thus, if the prediction system shows a winrate of

80% or higher in a best of 1
90% or higher in a best of 3
94% or higher in a best of 5
96% or higher in a best of 7
the numbers should be taken with a grain of salt.

On April 19 2013 17:52 Clazziquai10 wrote:
Lucifron > Bogus, MKP, Maru and all those other sick korean ballers made me lol XD