Checking if Elo system is oppressive [With proofs]

Sort:
IndianCamels

Your model is using glicko

basketstorm
IndianCamels wrote:
basketstorm wrote:
IndianCamels wrote:

There are more reasons why this is inaccurate. If we were using elo, the higher the rating of the player, the less players of his rating, the less points he gets per game. This is why it is so hard for the top 10 to hit 2900. Simply put, they would have to win many chess games in a row.

That's not how it works. You don't get less points per game just because there are not enough players in your rating range.

Look up the calculation for elo and then tell me.

Tell you what? I know how Elo works and I know that for 85% chance difference must be 301.33107, not 400. Do you know why it's 301.33107 and not 300 and not 400?
Glicko is based on Elo, it just defines how ratings changes apply. Deviation accounts for inactivity, but it doesn't take long for the system to deem your rating "accurate" and all actual inaccuracies are then unfairly reflected on the ratings.

abrilxinho

.

basketstorm

If anyone is interested in proving that Elo rating is NOT oppressive, show in math, formulas, simulations etc HOW a non-oppressive Elo system is possible.
No need to explain how Elo is supposed to work. That's clear. Let's think how we can actually make it work. 
It's not about definition of Elo it's about how it is assigned and changed for everyone.And please if you can't at least roughly outline it with simulation you don't even understand how it works, then I see no point in your denial.

Take just two players.Both just signed up. Easy, right?Both have 200. With 100 as rating floor.First player is stronger and wins 75% of the games.

Run 100, run even 1000 games between them. Their rating will almost never reflect that 0.75 probability.

Results after 1000 games:

Weaker - 102

Stronger - 430

Elo difference = 328

Another 1000-game run from scratch:

Weaker - 185

Stronger - 344

Elo difference = 159

But Elo is defined in such way that they should have 191 Elo difference if one wins 75% of the time.

Add a third player to the equation. He is stronger than those two. He wins against stronger player 90% of the time.Theoretical rating difference should be 382. They start as 430 and 200 rated players. But after a 1000 games, typical result looks like this: one has 106, other 626. (Difference: 520)

Elo strength of these players can be expressed as: 1000, 1191, 1573. Or add 400 to each: 1400, 1591, 1973.

But their assigned rating as result of games is completely chaotic, in favor of the strongest player of course.

If you're unlucky enough to meet a player that is stronger than you but has same rating, your rating will unfairly drop. And if you were underrated before because of such drops, you would be even more underrated. And to some other player who is weaker than you but has same rating as your new rating it would be also an unfair game, or rather an unfair rating exchange. And it's not enough to play 100 or even 1000 games to balance this out.

IndianCamels

Indeed, the creator of elo himself said "its about your performance

xtreme2020
#58 the definition of elo is your skill. If you understand how elo works, you understand that it gets you to your skill
xtreme2020
#51 no, I should have put “crazy rating mismatch” in quotes
IndianCamels

Oppressive is the wrong word. Oppressive would be actively working against one to prevent their success. It's not oppressive. You and I have different skill sets, training levels, life experiences... These all play a part in the result of the game. However, the rating system works the same for us. If it truly was oppressive, then a 400 rapid should equal 2000 rapid, which is clearly not the case. Every 200 points, there is a clear skill distinction.

xtreme2020
To logically prove it: if your skill level is better than others, you will win more than 50% of the time. If it’s equal, you’ll win 50% of the time. Elo is designed so that at a certain point you reach a point where you win 50% of your games. This means you’re playing players of your same level. Whatever number you’re at then, is your true skill level compared to everyone else’s
basketstorm
xtreme2020 wrote:
#58 the definition of elo is your skill. If you understand how elo works, you understand that it gets you to your skill

No, it's not skill. It does measure skill difference but mainly serves as further win probability prediction. Elo difference directly translates to win% and back.

Quote: Elo ratings are comparative only, and are valid only within the rating pool in which they were calculated, rather than being an absolute measure of a player's strength. 

But, how to assign the rating and how to change it, what K-factor to use? And how to adjust to skill growth? This becomes messy quickly.

xtreme2020
Yes, of course elo is comparative, however there aren’t really separated “rating pools”, just one which included everyone. It doesn’t really get messy, as the system now works fine, there’s not much to it
xtreme2020
Separate rating pools would require them to be completely isolated, however anyone can go up any amount of rating and same down, and they do. There’s only one rating pool, which includes everyone
xtreme2020
Skill difference is the same thing as saying skill. If I have a 50% win probability against someone, I am the same skill. If the other person has a better win probability, they’re better, and vice versa. This is why elo=skill
xtreme2020
This is just janko if he knew a tiny bit of English lol
IndianCamels

Gatekeeping isn't real in chess.com

TheCobraisaready

Has there ever been a game where each move was identical to a famous match? What are the probability of that happening, would it have to an infinite number of games before that would happen?

MasterJyanM
TheCobraisaready wrote:

Has there ever been a game where each move was identical to a famous match? What are the probability of that happening, would it have to an infinite number of games before that would happen?

easy,one in a one;take a game and copy it

TheCobraisaready
MasterJyanM wrote:
TheCobraisaready wrote:

Has there ever been a game where each move was identical to a famous match? What are the probability of that happening, would it have to an infinite number of games before that would happen?

easy,one in a one;take a game and copy it

Nah bro I mean if you say took a million games played simultaneously how many would make identical moves like the whole game would be identical without knowing it or copying, what is the probability of that happening? that's rhetorical

MasterJyanM
TheCobraisaready wrote:
MasterJyanM wrote:
TheCobraisaready wrote:

Has there ever been a game where each move was identical to a famous match? What are the probability of that happening, would it have to an infinite number of games before that would happen?

easy,one in a one;take a game and copy it

Nah bro I mean if you say took a million games played simultaneously how many would make identical moves like the whole game would be identical without knowing it or copying, what is the probability of that happening? that's rhetorical

you would have to use the best of the best engiene moves and a random game and compare them,seeing which moves are identical.repeat this with other random games until you have reached a completely or mostly complete identical game.

basketstorm
xtreme2020 wrote:
To logically prove it: if your skill level is better than others, you will win more than 50% of the time. If it’s equal, you’ll win 50% of the time. Elo is designed so that at a certain point you reach a point where you win 50% of your games. This means you’re playing players of your same level. Whatever number you’re at then, is your true skill level compared to everyone else’s

But pairing isn't fair and penalty isn't fair either. If 200 beats you as 800 (because he is underrated), it's not fair to lose 15 points. You get 50% on average because some of your opponents are overrated while some are underrated, not because their skill matches your skill.

Rating floor and ridiculously small initial rating are the major issues. I have already demonstrated this in 2-player example.

If you remove the floor, you'll get negative ratings and if 200 is your starting rating, result will look like this:

But at least:
- there would be no local inflation
- sorting will go on even in low-Elo area without turning into random chaos like now

Many players are actually there, "under the ground", below zero, I mean they should've had negative rating. They should have their place, their crowns and thrones in the underground chess federation. Everyone is better than others (except for some lone lowest ultra-beginner). We are not equal. But oppressive chess matrix refuses to admit our strengths. It just clips the rating at 100. Keeps us close to that edge. It randomly nudges us up and down with random strength because it thinks ratings are "accurate enough". What is 100, what is 200, what is 400? Engines in that setting are mere monkeys. Why did they squeeze the range of human chess skills into something so thin and small?
Why am I facing 127-rated players that are either very weak or beat me easily? Because no one knows how many times they've hit the floor.

Don't want negative rating - make starting rating larger. Much larger. But it doesn't matter what is the starting value as long as there's no clipping and movement up or down is fair.

Want to measure actual skill - measure against a precisely set up engine.

You can't expect to have a single pool especially when the playerbase is that large. In reality it's always a web of local, isolated pools. And largest web of webs is that low-Elo web as shown in chess.com leaderboard. Because of its chaotic nature (random opponent's skill, low starting rating and the rating floor) and unfair penalties, I think the word "oppressing" fits very well there.