Home » Forum Home » General

Topic: Draws in ratings
Replies: 20   Views: 163,005   Pages: 2   Last Post: May 3, 2019, 5:02 PM by: haggis

Search Forum

Back to Topic List Topics: [ Previous | Next ]
Replies: 20   Views: 163,005   Pages: 2   [ 1 2 | Next ]
rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Draws in ratings
Posted: Apr 27, 2019, 2:47 PM

Starting next week, somewhere on Monday April 29th, the ratings system will be altered to account for draws. I have already updated the FAQ to reflect the upcoming changes.

For more background, a lengthy discussion can be found here


watsu

Posts: 1,461
Registered: Dec 16, 2001
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 3:52 PM

Will there be a rating floor implemented as haijinx describes is in place with chess?

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 3:54 PM

IIRC, a rating floor is permanent, and can only go up.

I am questioning the reasoning behind using it before ratings have settled after including draws in rating adjustments. If ratings are indeed inflated as claimed, isn't adding a floor counterproductive?

Adding floors gives opportunity for cheating here, even if you wanted to use a floor that was achieved after the rating became established, someone could create a few new accounts, inflate their rating until it became established and then roll on with a high floor.

I can see floors working fairly well in a curated world like the USCF, but there's hardly curation going on here. I am speculating that adding rating floors, especially right when the formula is adjusted, is premature.


Message was edited by: rainwolf at Apr 27, 2019 4:03 PM

watsu

Posts: 1,461
Registered: Dec 16, 2001
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 5:08 PM

My prediction is that with this implementation ratings will compress towards a lower value than the top ratings. If a 2300 player can split 3 sets with a 2600 player ( not unrealistic) and yet also split three sets with a 2000 player (also not unrealistic) then 26 goes down 23 stays about the same and 2000 goes up. The floor helps prevent 2600 from going below 2400. Players like Zk, nosovs and Richiii who haven't played much TB here will move up to the new ceiling should they choose, but the new ceiling will be lower than the old one, so the floor doesn't benefit them, it just mitigates compression down from the current top.

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 5:27 PM

This feels like 2 measures that counter themselves a bit.
I'm not sure I can pull this off before Monday and might have to postpone it all then.

For rating floors, should only established ratings be considered?

watsu

Posts: 1,461
Registered: Dec 16, 2001
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 6:15 PM

The floor does counter some of the likely effect of the draws counting at the top level, but shouldn't be much of a factor otherwise at other levels, as points from the top get redistributed downward.
I don't know if the floor matters for provisional calcs.

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 6:26 PM

I'm postponing it for now, adding the floors has made it considerably more complex. Not for future calculation, but finding the floors for all users if I exclude the provisional games. In which case I'd have to ignore the 20 first wins/losses.

watsu

Posts: 1,461
Registered: Dec 16, 2001
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 7:55 PM

I think a bit more notice is not a bad thing in any case, since TB games can last weeks at times.

Retired from TB Pente, but still playing live games & exploring variants like D, poof and boat
rainwolf

Posts: 763
Registered: Apr 12, 2008
From: Singapore
Age: 44
Home page
Re: Draws in ratings
Posted: Apr 27, 2019, 9:24 PM

I'll initialise everyone's rating floor based on the last rating they had before the rule came into effect, and only start considering floors once ratings become established.

xenophon

Posts: 19
Registered: Apr 19, 2011
From: Terra
Re: Draws in ratings
Posted: Apr 28, 2019, 10:59 AM

Every game played rationally should end in a draw. Moving first or second should result in a (dis)advantage which is reversed when the order is reversed. As such, any set should result in a draw. Draws are the default, not the aberration. Smarter players will be disincentivized from playing lower ranking players.

If you lose you should lose rank, a risk we all take. If you win, you gain rank. Drawing can result in the gaining or losing of rank based on the relative rank of your opponent. This will serve as a disincentive to play against players of lower ranking.

karlw

Posts: 973
Registered: Mar 7, 2006
From: Eugene, Oregon
Age: 36
Re: Draws in ratings
Posted: Apr 29, 2019, 3:26 AM

> Every game played rationally should end in a draw.

This is false. Assuming we are speaking about sets, a set played between a beginner and a master should not end in a draw. A beginner must make at least one serious mistake (otherwise he isn't a beginner), and all the master needs to do is capitalize on that mistake and make no mistakes of his own (as masters must do if they want to call themselves masters). Rationally, a set between a beginner and a master should end in favor of the master 100% of the time.

Returning to the debate about whether draws should be rated, I encourage you to read the many posts by rainwolf, haijinx, watsu and myself where we discuss the pros and cons in detail. To summarize the arguments presented in favor:

-The creator of the Elo system believed that draws should be rated and provided a method by which to do so. Having a large proportion of rated games not be rated is illogical and may lead to systematic bias (see below).

-Immediately following the change to set-based play at pente.org (and not rating draws), rating inflation became a serious issue, to the point that ~1800 in game-based play equates to ~2300 (or higher) in set-based play.

-This system does not disincentivize high-ranked players from playing low-ranked players. But this itself is something of a problem, because they are able to play essentially risk-free. An intermediate player might occasionally be able to take a game from an expert (and should be rewarded for doing so in the ratings), but to win two in a row is a Herculean task. Higher-rated players can farm lower-rated players for points, which leads to the inflation discussed above.

It's no use going back to yesterday, because I was a different person then.
valleyboy

Posts: 121
Registered: Mar 2, 2005
From: McAllen TX
Age: 56
Re: Draws in ratings
Posted: Apr 29, 2019, 3:34 AM

This is something I asked about in a forum post within the past year, was told that it would require some "knotty algorithms" (which I just had to look up the spelling for).

Higher, sometimes much higher, rated players sometimes accept my open invitations...and I groan...nothing to gain and often too far above my pea brain ability to learn from. NOW I may have something to look forward to in that circumstance!

And as it is bad form to refuse or ignore an invitation, we may see the ratings tightening, the "any given Sunday" syndrome at work.

haijinx

Posts: 64
Registered: Jan 20, 2019
From: Salem Oregon
Age: 48
Home page
Re: Draws in ratings
Posted: Apr 29, 2019, 4:59 AM

About the higher rated players won't play lower players fallacy...

Obviously, since we're not all equally rated, virtually 99.9999% of the games are between a higher and lower rated player. Given that, I feel very safe to wager that games will still occur here and probably at about the same frequency of games/active players.

Only the ratings will be a closer approximation of relative playing strength once the system properly accounts for all games and ratings contract.

Look at it this way, if I draw every set with a player...we both have P1 games that are shiny for example...shouldn't our ratings converge over time to match the probability of us winning at 50/50...depending on who goes first?

Well, if draws are ignored, that never happens leading to inflation. I'm glad it's being fixed and I think that rating floors is a good safety net for those who are vested in their rating class.

snapding

Posts: 7
Registered: Mar 23, 2004
From: Michigan
Age: 41
Re: Draws in ratings
Posted: May 1, 2019, 5:32 AM

My two cents. (Which might count as one) any time i draw against a super player like watsu (and a few others) I feel like it's a win. I feel like it should count for something. The warm fuzzy feeling is good enough for me, but maybe a rating boost for achieving what seems like the impossible would feel warmer and fuzzier.
On this note I dont really think I should be rated in the same category as players like watsu (I'm singling you out because i have never won a game against you... ever) this might be a topic for another discussion but maybe a new category is in order. I dont consider myself a master. Maybe submaster or something in that order.

pente_gon

Posts: 54
Registered: Dec 26, 2017
Re: Draws in ratings
Posted: May 1, 2019, 10:51 AM

I agree that draws should be take into account but in my opinion rating floors are not necessary - it is not natural. And for sure not now, because rating system is (was) unstable.

For example my rating is to high and it was possible only because draws were not taken into account. It should go down and after some time should reach "stable" level.

Replies: 20   Views: 163,005   Pages: 2   [ 1 2 | Next ]
Back to Topic List
Topics: [ Previous | Next ]


Powered by Jive Software