7

Is there any relation or any attempt at discovering a relation between Chess960 ratings and standard chess ratings? This can be for GMs, for (any)M's, for pro players, or for all players on lichess/chess.com (for whatever lichess/chess.com statistics are available), but preferably for as many categories/divisions/what you call it as possible.

I'm looking for something similar to this reddit post: I made a website to compare chess ratings between Chess.com, LiChess, FIDE, & USCF. This way 1 might see eg 9LX ratings on lichess are about 100 points less than standard blitz ratings on lichess or 200 points less than standard rapid ratings on lichess.

Something I thought: I guess there's weekly rating distributions.

For example in the 'week' (whatever this means) of 2021-11-12 13:50:07Z (the time now) 49% is 1575 in 9LX, 50% is 1525 in blitz and 49% is 1525 in rapid. Assuming everyone on lichess plays 9LX, blitz and rapid about the same, I guess this suggests 9LX ratings are about 50 points higher than regular ratings, which of course is not the case.

I guess what would be helpful is seeing the blitz and rapid percentiles conditioned on non-provisional 9LX ratings. Of course there's the lack of time control separation in 9LX ratings. Perhaps we might do some weighted average like if 90% 9LX games are blitz and 10% are rapid then.......idk.

standard:

enter image description here

9LX:

enter image description here

Related:

What are some solutions to a possible/perceived underratedness problem in online chess960?

Improve this question
6
  • 1
    @BCLC A good question. Personally I've often felt curious how strong this relationship is top GMs, and in particular who the strongest 960 players (now or ever) have been.
    – Mobeus Zoom
    Jan 20, 2021 at 16:29
  • 1
    It does seem to be clear that being good at standard chess is highly linked to being good at Chess960; even online tournaments which began open to everyone, or say, Titled players, have generally ended up seeing the familiar faces qualify (en.wikipedia.org/wiki/…) and the most familiar faces win (Wesley So and Magnus Carlsen).
    – Mobeus Zoom
    Jan 20, 2021 at 16:32
  • @MobeusZoom updated question. maybe rating distributions?
    – BCLC
    Nov 18, 2021 at 15:30
  • 3
    If you want the relation between Chess960 and standard chess ratings, overall rating distributions will be useless. You need to compare the standard chess and Chess960 ratings of each player (individually) and plot them on a chart, make a model to predict one from the other, etc. My breath is most likely wasted no doubt - as often, your questions are straightforward to answer but not without the grunt work (querying Lichess or Chess.com ratings for a large number of users), which 50 points of SE reputation is unlikely to even begin to approach paying for, and which you're unlikely to put in.
    – Mobeus Zoom
    Nov 18, 2021 at 21:35
  • 1
    "which of course is not the case" I'll add we didn't say this. It is possible that the median Chess960 rating is lower or higher than the median standard rating on Lichess (cf e.g. chess.stackexchange.com/questions/37352/…). This does not mean anyone is "underrated".
    – Mobeus Zoom
    Nov 18, 2021 at 22:07

1 Answer 1

Reset to default
5
+200

Web app called rating correlations is now published in streamlit cloud. It is slow so just be patient, will try to optimize it later.

It can also estimate crazyhouse aside from chess960. Rating data are all from Lichesss.

You can input by double-clicking the number and type your bullet rating for example, the -/+ button is not necessary.

App is updated for performance and source code can be found in github.

Update 1: Can estimate rating based from chess.com users.

enter image description here

url: https://ratingcorrelations.herokuapp.com

Improve this answer
6
  • Look forward to the publication. Thanks. So far this red line does confirm some of my suspicions
    – BCLC
    Jan 23, 2022 at 6:29
  • ferdy, can you say anything similar to 'A rating of 1600 in LiChess.org: Blitz would equal 1759 in LiChess.org: Rapid, with an average variation of plus/minus of 101 points' in chessratingcomparison.com/graphs ?
    – BCLC
    Jan 23, 2022 at 6:35
  • 1
    I got 1739 for that 1600.
    – ferdy
    Jan 23, 2022 at 7:17
  • right thanks. i mean in general, does your web app have this kind of thing similar to what chessratingcomparison.com has ( 'A rating of 1600 in LiChess.org: Blitz would equal 1759 in LiChess.org: Rapid, with an average variation of plus/minus of 101 points')? but of course your web app in progress is really incredible already
    – BCLC
    Jan 23, 2022 at 15:23
  • 1
    From current UI, change feature to Blitz, then change target to Rapid, input 1600 in the conversion container then estimated Rapid rating will be generated. I have not implemented yet the margin of error. Currently using xgboost as regressor. But I am also looking on statsmodels. I will compare its performance and speed especially when doing multiple regression like bullet + blitz + rapid to predict chess960 for example.
    – ferdy
    Jan 23, 2022 at 16:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.