As usual, i completely disagree, you do not understand what a proper ranking system should look like.
If poker is not the skill based game, lol, why do you not go all in all the time?

30 games does not reflect the skill at all, even 100 games is meaningless
if the system is not skill based.

Beating average pokerth player should be worth almost nothing but beating someone
good (better than you) should be worth a lot. And every game has an impact!
The bigger the win, the higher increase or ranking. The same goes for loss ...

Currently the best strategy is to play against skillless players.
Enjoy your the current meaningless ranking. If you do not trust your skill, i can see your point.

Glicko-2 rating is even better than elo, only tweaking of paramaters left to make it appropriate for poker and our WEC.

Cheers, filer.

gfgh wrote:
Nothing is a 100% skill game. And Elo doesn't assume anything to that effect. Much to the contrary - if better players would always beat worse players, Elo scores would go to infinity, making them completely wortheless.

The point of Elo is to approximate winning propability. Even if the best player had only 55% win rate against the worst player - given enough time, Elo would tease out that exact fact.
gfgh wrote:
That's an interesting idea. I suppose you could, say, try to measure whether an opponent was playing aggressively or passively - and then award each player two (or more) Elo scores playing against each style. Wouldn't be too hard to built such a system.

I'm not really sure it's worth it though. Personally, I think you should only be allowed to call yourself a good poker player once you can win against any style out there. Either by adopting a save style yourself - or by reacting to your opponent's play.

I have my implementations of glicko2 almost done. But this weighting is a pickle. Do you have any other idea?
I do not like linearity too much.
How about that? We compute the rating deviance of a game
as \sqrt{ \sum_{ij} (S_i-S_j)^2} and then apply weights according to that?
Higher variance more from all one weighting we should be.

But when i think about, that probably over complicates.

Cheers, filer.

filer wrote:
I'm not really a Glicko expert by any means... Is there a good reference somwhere to how it is derived? There must be a way in there to represent less relevant results.
filer wrote:
Not sure what you mean?

Your proposed way of weighting is more or less linear in the log scale?

Here is the complete glicko-2 reference, see the pdf on the site
www.glicko.net/glicko.html

If you are interested in Math background
math.bu.edu/people/mg/research/glicko.pdf

Cheers, filer.

Well, the weights are obviously just guesswork. It would be better to use some statistical data from real Poker matches about how many hands someone typically plays when placing Nth place in the end.

And yeah, I had seen that site. But the PDFs linked don't explain much about where the formulas come from or what they mean. It's hard to make a call at what position the formulas need to be changed this way... There's the "Parameter estimation in large dynamic paired comparison experiment" paper, but that describes the Glicko-1 system, right?