First knight odds match with IM Saturday

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
todd
Posts: 56
Joined: Thu Apr 19, 2018 7:09 pm

Re: First knight odds match with IM Saturday

Post by todd » Thu Jan 09, 2020 8:52 am

The lichess rating range is compressed relative to chess.com, with lichess ratings being higher for the vast majority of players and lower at the very top.

I'm not sure exactly where the break-even point is, but somewhere in the 2300s is a reasonable guess - I myself spend most of my time in the 2300s blitz on both sites.

lkaufman
Posts: 4218
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: First knight odds match with IM Saturday

Post by lkaufman » Thu Jan 09, 2020 6:00 pm

todd wrote:
Thu Jan 09, 2020 8:52 am
The lichess rating range is compressed relative to chess.com, with lichess ratings being higher for the vast majority of players and lower at the very top.

I'm not sure exactly where the break-even point is, but somewhere in the 2300s is a reasonable guess - I myself spend most of my time in the 2300s blitz on both sites.
Thanks, that is quite helpful. Any idea why lichess range is compressed relative to chess.com? Some difference either in the rating formula or how pairings are done?
Komodo rules!

User avatar
Ovyron
Posts: 4241
Joined: Tue Jul 03, 2007 2:30 am

Re: First knight odds match with IM Saturday

Post by Ovyron » Fri Jan 10, 2020 1:36 am

One thing I can say for certain is that chess.com is flooded by the weakest chess players on the entire internet, perhaps the artifacts would disappear if lichess' player pool was similarly flooded by so weak players.

I remember when I started joining chess.com tourneys and was amazed at how I was beating 4 people in a row and easily staying in first place, before figuring what was going on. Perhaps this is related.

lkaufman
Posts: 4218
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: First knight odds match with IM Saturday

Post by lkaufman » Fri Jan 10, 2020 2:11 am

Ovyron wrote:
Fri Jan 10, 2020 1:36 am
One thing I can say for certain is that chess.com is flooded by the weakest chess players on the entire internet, perhaps the artifacts would disappear if lichess' player pool was similarly flooded by so weak players.

I remember when I started joining chess.com tourneys and was amazed at how I was beating 4 people in a row and easily staying in first place, before figuring what was going on. Perhaps this is related.
Well, I can see why that would cause an average tournament player (say 1500 elo or so) to have a higher rating on chess.com than on lichess, but that's not what we're told is happening. Apparently such players have LOWER ratings on chess.com than on lichess. So that doesn't seem like a plausible explanation.
Komodo rules!

todd
Posts: 56
Joined: Thu Apr 19, 2018 7:09 pm

Re: First knight odds match with IM Saturday

Post by todd » Fri Jan 10, 2020 4:26 pm

lkaufman wrote:
Thu Jan 09, 2020 6:00 pm
todd wrote:
Thu Jan 09, 2020 8:52 am
The lichess rating range is compressed relative to chess.com, with lichess ratings being higher for the vast majority of players and lower at the very top.

I'm not sure exactly where the break-even point is, but somewhere in the 2300s is a reasonable guess - I myself spend most of my time in the 2300s blitz on both sites.
Thanks, that is quite helpful. Any idea why lichess range is compressed relative to chess.com? Some difference either in the rating formula or how pairings are done?
Here are graphs of the rating distributions at each site:
https://lichess.org/stat/rating/distribution/blitz
https://www.chess.com/leaderboard/live (the graph is in the upper right and may not show if you are not logged in)

There are differences in the rating formula, but I am not sure how those differences translate to systemic differences in ratings or whether one system will tend to be more or less inflationary or deflationary over time.

A partial (but probably not complete) explanation can be given by a few things:

1) Lichess has a rating floor that for a long time was at 800 (recently lowered to 600) and chess.com goes down to 100. Some tournament and other serious players may not be aware, but there are a HUGE number of casual chess players and beginners at this level. The sort of people who end up playing in tournaments or having chess as a major lifelong hobby tend to breeze past 1000 quickly, but we are a minority, and most people do not. The average chess.com blitz player is rated 989 (see the link above to verify for yourself).

2) Strong players on lichess "berzerk" in tournaments, which lowers their ratings. Berzerk cuts your time in half (and forces you to play without increment if the tournament has increment) but gives you extra tournament points for winning.

3) The initial rating for new players is higher on lichess (on chess.com you can choose a higher initial rating, but it's not the default option)
Last edited by todd on Fri Jan 10, 2020 4:49 pm, edited 1 time in total.

lkaufman
Posts: 4218
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: First knight odds match with IM Saturday

Post by lkaufman » Fri Jan 10, 2020 4:32 pm

todd wrote:
Fri Jan 10, 2020 4:26 pm
lkaufman wrote:
Thu Jan 09, 2020 6:00 pm
todd wrote:
Thu Jan 09, 2020 8:52 am
The lichess rating range is compressed relative to chess.com, with lichess ratings being higher for the vast majority of players and lower at the very top.

I'm not sure exactly where the break-even point is, but somewhere in the 2300s is a reasonable guess - I myself spend most of my time in the 2300s blitz on both sites.
Thanks, that is quite helpful. Any idea why lichess range is compressed relative to chess.com? Some difference either in the rating formula or how pairings are done?
Here are graphs of the rating distributions at each site:
https://lichess.org/stat/rating/distribution/blitz
https://www.chess.com/leaderboard/live

There are differences in the rating formula, but I am not sure how those differences translate to systemic differences in ratings or whether one system will tend to be more or less inflationary or deflationary over time.

A partial (but probably not complete) explanation can be given by a few things:

1) Lichess has a rating floor that for a long time was at 800 (recently lowered to 600) and chess.com goes down to 100. Some tournament and other serious players may not be aware, but there are a HUGE number of casual chess players and beginners at this level. The sort of people who end up playing in tournaments or having chess as a major lifelong hobby tend to breeze past 1000 quickly, but we are a minority, and most people do not. The average chess.com blitz player is rated 989 (see the link above to verify for yourself).

2) Strong players on lichess "berzerk" in tournaments, which lowers their ratings. Berzerk cuts your time in half (and forces you to play without increment if the tournament has increment) but gives you extra tournament points for winning.

3) The initial rating for new players is higher on lichess (on chess.com you can choose a higher initial rating, but it's not the default option)
This sounds like a full explanation. The Berzerk option would totally trash the expectancy formula and would indeed contract the higher ratings. It makes the math behind the rating system invalid. Basically they are rating handicap games as if they were normal games. What a bad idea!
Komodo rules!

User avatar
Ovyron
Posts: 4241
Joined: Tue Jul 03, 2007 2:30 am

Re: First knight odds match with IM Saturday

Post by Ovyron » Fri Jan 10, 2020 5:14 pm

The idea is great, remember how you were saying the other day that if you were invited to a tournament, first you'd want to see who else was participating, and if they all were some 300 elo points weaker than you, you'd not play, because you'd only be interested in playing if there was some challenge?

Well, this Berzerk option solves just that, you cut your time in half, suddenly your opponent that was going to be very easy to beat becomes a challenge, and the game becomes exciting again as your clock starts running out!

It also adds an extra dimension to the game, because, if you check games between mismatched players, you'll see that in most of them the stronger player didn't use even half their clock, so what use did the extra time have? None, so players have the option to exchange it for an extra point if they win (in this system there's 1 point for a draw and 2 for a win, so by berzerk winning you get one win and a draw worth of points.)

The question is, when should you Berserk? You should only berserk when you're certain that you will beat this opponent you're facing, and that you will do it with more than half your time remaining, then berserk or no berserk makes no difference, except that you get an extra point after the game is over.

This means that there's absolutely nothing wrong with the Berserk system, and any rating disruption that occurs is the fault of the players for Berserking when they shouldn't, because they assume they'll be able to defeat an opponent quickly but they can't.

Turns out the only ratings that are contracted at the top are from those that haven't learned how to berserk properly, but this makes lichess tournaments the most exciting ones on the Internet!

User avatar
AdminX
Posts: 5426
Joined: Mon Mar 13, 2006 1:34 pm
Location: Acworth, GA
Contact:

Re: First knight odds match with IM Saturday

Post by AdminX » Fri Jan 10, 2020 6:00 pm

todd wrote:
Fri Jan 10, 2020 4:26 pm
Here are graphs of the rating distributions at each site:
https://lichess.org/stat/rating/distribution/blitz
Thanks for sharing this, LiChess has so many statistical and analytical tools, it really more than I can say grace over. Discover something new over there on a regular basis. :D
"Good decisions come from experience, and experience comes from bad decisions."
__________________________________________________________________
Ted Summers

lkaufman
Posts: 4218
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: First knight odds match with IM Saturday

Post by lkaufman » Fri Jan 10, 2020 6:40 pm

Ovyron wrote:
Fri Jan 10, 2020 5:14 pm
The idea is great, remember how you were saying the other day that if you were invited to a tournament, first you'd want to see who else was participating, and if they all were some 300 elo points weaker than you, you'd not play, because you'd only be interested in playing if there was some challenge?

Well, this Berzerk option solves just that, you cut your time in half, suddenly your opponent that was going to be very easy to beat becomes a challenge, and the game becomes exciting again as your clock starts running out!

It also adds an extra dimension to the game, because, if you check games between mismatched players, you'll see that in most of them the stronger player didn't use even half their clock, so what use did the extra time have? None, so players have the option to exchange it for an extra point if they win (in this system there's 1 point for a draw and 2 for a win, so by berzerk winning you get one win and a draw worth of points.)

The question is, when should you Berserk? You should only berserk when you're certain that you will beat this opponent you're facing, and that you will do it with more than half your time remaining, then berserk or no berserk makes no difference, except that you get an extra point after the game is over.

This means that there's absolutely nothing wrong with the Berserk system, and any rating disruption that occurs is the fault of the players for Berserking when they shouldn't, because they assume they'll be able to defeat an opponent quickly but they can't.

Turns out the only ratings that are contracted at the top are from those that haven't learned how to berserk properly, but this makes lichess tournaments the most exciting ones on the Internet!
Your argument has a flaw. If your goal is to win the tournament, then you should berserk any time your chance of winning in half the time is very high (maybe 90% or so, just guessing), not only when it is 100%. But when you do this against players whom you should score 90% against, it may lower your win exp. to say 85%, so you will consistently lose rating points even though it is the proper strategy for the tournament. This is a problem any time tournament prizes are based on anything different than the scores used by the rating formula. It is for this reason that tournaments with "football" scoring (0 loss, 1 draw, 3 win) should really not be ratable by FIDE, although in practice the distortion is small so they have ignored this issue. The berzerk problem seems more serious to me than the football scoring problem.
Komodo rules!

User avatar
Ovyron
Posts: 4241
Joined: Tue Jul 03, 2007 2:30 am

Re: First knight odds match with IM Saturday

Post by Ovyron » Fri Jan 10, 2020 7:01 pm

lkaufman wrote:
Fri Jan 10, 2020 6:40 pm
Your argument has a flaw. If your goal is to win the tournament, then you should berserk any time your chance of winning in half the time is very high (maybe 90% or so, just guessing)
Not so :) I've watched over 60 hours of videos of people Berserking, and different Berserking strategies. You should only Berserk when you're certain of a win.

This is because of another factor happening in there: wins in a row bonuses. Once your winning streak hits 2, the next game you win will give you 4 points (and now wins will be worth 4 points), but if you draw or lose you'll go back to normal scoring, and will have to win again 2 on a row to be on a streak, which means overall you'll get more points without risk if you continue winning games than if you get Berserk points, so at this point if you Berserk you win 5 points vs 4 (not worth the risk) and it's better to not Berserk (unless you'll win anyway).

Winning games in a row is more profitable than Berserking, so the more you win, there more that is at stake if you lose a game, even if you just have 10% losing chance, it's not worth the risk.

Post Reply