Your network blocks the Lichess assets!

lichess.org
Donate

FIDE rating changes: Are they working so far?

Nice analysis Vlad

Nice analysis Vlad

Very cool visualization, the 'honey comb' (?) graph!

Very cool visualization, the 'honey comb' (?) graph!

It's helping more people climb to 2300 so they can pay for a title. That's what FIDE intended with the change.

It's helping more people climb to 2300 so they can pay for a title. That's what FIDE intended with the change.

URS is way better obviouly. Just like any maximum likelyhood system that uses also history. System like glicko and elo are both badly dated. But URS has for online several problems one technical several psychological ones

  • Any Max-likelihood system consumes way more as it need find a minimum for really big multi variable function
  • Instant gratification is what internet is for. People want see the change in rating at once. Which obviously wont happen as all ML algorithms are heavy so they are calculated periodically - period can be short like 15 second or a minute. Depending heavily on available resources.
  • People by and large fail to understand that rating points are not money like property but just number that tries to be accurate estimate of rating. All ML systems can change you rating even if you don't play if indirect evidence of your strength comes available. And quite some people find that infuriating.

That said almost all Go-servers have used some sort full history maximum likelihood rating system from day one. So at Go players have managed to live with them

Nice article and shows that tiny patchin of Elo will not help too much.

URS is way better obviouly. Just like any maximum likelyhood system that uses also history. System like glicko and elo are both badly dated. But URS has for online several problems one technical several psychological ones - Any Max-likelihood system consumes way more as it need find a minimum for really big multi variable function - Instant gratification is what internet is for. People want see the change in rating at once. Which obviously wont happen as all ML algorithms are heavy so they are calculated periodically - period can be short like 15 second or a minute. Depending heavily on available resources. - People by and large fail to understand that rating points are not money like property but just number that tries to be accurate estimate of rating. All ML systems can change you rating even if you don't play if indirect evidence of your strength comes available. And quite some people find that infuriating. That said almost all Go-servers have used some sort full history maximum likelihood rating system from day one. So at Go players have managed to live with them Nice article and shows that tiny patchin of Elo will not help too much.

Very interesting article! One thing I wonder is how much the new ratings have affected players rated ~2100. I think there might be a possibility this change wasn't just "for the greater good of chess" but also to inflate the elo of players around that level so there will be more players that get to 2200-2300 level, meaning they will recieve titles. In case you guys didn't know, FIDE charges a price to get a title, meaning the more titles the more revenue for them. However, this is just a theory and I'd like to see how the statistics shows this (and maybe it can be calculated exactly how much they made).

Very interesting article! One thing I wonder is how much the new ratings have affected players rated ~2100. I think there might be a possibility this change wasn't just "for the greater good of chess" but also to inflate the elo of players around that level so there will be more players that get to 2200-2300 level, meaning they will recieve titles. In case you guys didn't know, FIDE charges a price to get a title, meaning the more titles the more revenue for them. However, this is just a theory and I'd like to see how the statistics shows this (and maybe it can be calculated exactly how much they made).

revenue from titles is negligible amount. FIDE this year budget is about 16 M€ and estimated title application fees is 150 000. It is money but tiny increases into is pocket money

revenue from titles is negligible amount. FIDE this year budget is about 16 M€ and estimated title application fees is 150 000. It is money but tiny increases into is pocket money

I do like the structured presentation. And the link to source files. Without having gone through them but just from the text, it seems to be offering a needed (for me) set of pointers to getting some basics about official chess circles rating systems, from the OTB cohorts. (:). Something I would need to study further as it seems to be offering a wide angle approach. Rating is not my cup of tea about chess topics and data analysis but they are all we have for now. And some chess board logic questions can be informed by understanding how to use ratings as predictor or probabality estimators. Win ratio, but how when and which context. I am learning. No need to reply to this. But thanks for this well introduced chunk of pointers.

I do like the structured presentation. And the link to source files. Without having gone through them but just from the text, it seems to be offering a needed (for me) set of pointers to getting some basics about official chess circles rating systems, from the OTB cohorts. (:). Something I would need to study further as it seems to be offering a wide angle approach. Rating is not my cup of tea about chess topics and data analysis but they are all we have for now. And some chess board logic questions can be informed by understanding how to use ratings as predictor or probabality estimators. Win ratio, but how when and which context. I am learning. No need to reply to this. But thanks for this well introduced chunk of pointers.

The problems is that the rated games distribution is by nature NOT normal distribution. People who is likely to win a match will be more likely to play it, and vice versa. Especially when the ticket fee and rewards are both high these days....

Elo will not work for such systems I fear.

Better game arrange system or continuing to Glicko probably will work...

The problems is that the rated games distribution is by nature NOT normal distribution. People who is likely to win a match will be more likely to play it, and vice versa. Especially when the ticket fee and rewards are both high these days.... Elo will not work for such systems I fear. Better game arrange system or continuing to Glicko probably will work...

This topic is now closed.