cant you give some interactive online version?

nice work anyway

It should not be that hard to pull off in fact. But my front-end skills leaves much to be desired :)

Ok, I'll do a backend if somebody wants to host it. But we need something UNIXy, preferably with around 2 Gig+ memory so we can keep everything in RAM and avoid disk seeks. Otherwise, we need a fast disk.

A G E N T
G
E
N
T

What about House of Decay?

I'll add it later to this thread :)

houseofdecay:

Rankings
Ladder

It would be really, really nice if you could compile the same information for lan only :)

I did that back in the day about a year ago. The problem is that there are relatively few LAN matches played and the winners get more matches than the losers of games, which skews the results a bit.

Online I have enough matches that the system works and does a good job at ranking players. Also, playing a lot does not hurt the ranking system as much if at all.

so basiclly...

agent best ql player ever?

I approve this ^^

Plus if you scrolled up and down a few times and made megusta face

I'm not incredibly volatile on dm13 it seems. Next up: AGENT

Plussed for incredibly awesome work as always god damnit.

Not volatility. If your horizontal bar is wide it means that the rating system has no confidence in the rating it has given you. If the bar is thin, it means that the system knows about your rating with greater confidence that it assigned the correct rating for you.

The confidence is affected by the opponent, the opponents rating and the confidence in the opponents rating. If you have a loss against a player with a slightly higher rating than you, but a much higher rating deviation (confidence) then it will not affect your rating and confidence that much. If on the other hand you have a loss against a player with better rating, and the deviation of the opponent is small, so we have a lot of confidence, then your rating and your rating deviation will be affected more.

One nice property of this is that if a good player creates a new account and then goes rampage on sorry n00bs, then this new account has so few matches that him winning won't really hurt the rating of the n00bs that much. Contrast with ELO/QLranks where this always hurts you a lot.

maybe you should put this in pdf...

x axis = ELO
dot = estimated skill
bar length = estimate boundaries

This is a good idea, perhaps except for the fact that the X axis is not ELO but Glicko 2 ratings :) I'll add it to the generator.

agentJkezooR and agent is a same player, data should be merged

I could do that, but it will take a bit of time since I do not want to mess with the raw data. Rather I have some derived data I can mess with. With 1.12 million matches, recalculating all matches actually takes an hour or so at the moment. But my code is not parallel even though I have 4 cores and 8 hyperthreads to work for me.

I'll write it down on the list of things to do. I think I know how to merge players in a good way though. It will hurt agents ratings though seeing that one player has a lower ranking than the other.