How is puzzle rating calculated?

Sort:
felpo4

Hello Everybody,

I'm trying to learn from tactics using puzzles. With the free membership I had a continuous improvement until 1300, then I noticed from a couple of day, that I'm strongly "punished" in case of error ( ie -12,-13 ) and in case of success a +1/+2. It seems to me that maybe something changed in the algorithm computing the rating, and it is making me converge to a new level. Is this possible? Anyone noticed such a behaviour as well?

notmtwain
felpo4 wrote:

Hello Everybody,

I'm trying to learn from tactics using puzzles. With the free membership I had a continuous improvement until 1300, then I noticed from a couple of day, that I'm strongly "punished" in case of error ( ie -12,-13 ) and in case of success a +1/+2. It seems to me that maybe something changed in the algorithm computing the rating, and it is making me converge to a new level. Is this possible? Anyone noticed such a behaviour as well?

The value of points diminishes with time.

There is a target time, based on your rating. If you solve the puzzle in less time, you get the full points.

There is also an average time. Again, if you solve the puzzle in less time, you get the most credit.

But if you get the first move of the puzzle wrong, you lose the full points. If you get one or two correct, you get partial credit.

USAuPzlBxBob

Yes… we're heavily penalized for incorrect moves.

Interesting, notmtwain's recent activity.

Clearly he's stuck at the 1200 level, and primarily because he is trying to get full credit by making his moves quickly.  I've been in that syndrome myself, around the 1700 — 1900 level, and then I decided I was going to just settle for 70% credit. (the take too long default credit level)

However, I take far too long on each puzzle.  But… I'm improving my puzzle strength solving ability, and I'm having good success, of late, in improving my score.  I just cracked the top 30,000 player level, which this morning stood at a score of 2228.

There are 990,022 of us working on our scores, with the number one position held by FlashyFerrari, who has a 95% success rate.  I don't know how he does it, but he has a YouTube I'll have to check out.

Here's a snapshot of his recent activity:

Date ID Rating Moves Target Time My Time Avg Time Outcome My Rating
Dec 28, 2019 0769346 3719 4/9 0:20 0:07 0:39 (44% | -114)
 
14743
Dec 28, 2019 0984482 3926 2/2 0:07 0:02 1:36 (100% | +1)
 
14857
Dec 28, 2019 0518534 3709 6/6 0:15 0:12 0:46 (100% | +1)
 
14856
Dec 28, 2019 0769192 3810 6/6 0:15 0:09 0:58 (100% | +1)
 
14855
Dec 28, 2019 0563236 3751 5/5 0:13 0:04 0:24 (100% | +1)
 
14854
Dec 28, 2019 0771278 3659 4/4 0:11 0:06 1:06 (100% | +1)
 
14853
Dec 28, 2019 0701898 3951 3/3 0:09 0:03 0:34 (100% | +1)
 
14852
Dec 28, 2019 1051808 3849 3/3 0:09 0:05 0:28 (100% | +1)
 
14851
Dec 28, 2019 0772230 3952 4/4 0:11 0:05 1:17 (100% | +1)
 
14850
Dec 28, 2019 0764704 3693 9/9 0:20 0:08 0:53 (100% | +1)
 
14849
Dec 28, 2019 0771058 3627 5/5 0:13 0:05 1:31 (100% | +1)
 
14848
Dec 28, 2019 0768390 3694 6/6 0:15 0:14 0:39 (100% | +1)
 
14847
Dec 28, 2019 0493538 3590 5/5 0:13 0:05 0:21 (100% | +1)
 
14846
Dec 28, 2019 0769364 3742 6/7 0:17 0:14 0:42 (86% | +1)
 
14845
Dec 28, 2019 0542682 3586 5/5 0:13 0:04 0:24 (100% | +1)
 
14844
Dec 28, 2019 0043181 3630 8/8 0:18 0:06 2:11 (100% | +1)
 
14843
Dec 28, 2019 0763926 3732 6/6 0:15 0:06 0:50 (100% | +1)
 
14842
Dec 28, 2019 0695776 3887 2/2 0:07 0:05 1:15 (100% | +1)
 
14841
Dec 26, 2019 0769192 3810 3/6 0:15 0:06 0:58 (50% | +1)
 
14840
Dec 26, 2019 0771278 3659 4/4 0:11 0:06 1:06 (100% | +1)
 
14839
Dec 26, 2019 0763926 3732 6/6 0:15 0:09 0:50 (100% | +1)
 
14838
Dec 26, 2019 0518534 3709 6/6 0:15 0:12 0:46 (100% | +1)
 
14837
Dec 26, 2019 0542682 3586 5/5 0:13 0:04 0:24 (100% | +1)
 
14836
Dec 26, 2019 0493538 3590 5/5 0:13 0:04 0:21 (100% | +1)
 
14835
Dec 26, 2019 0984482 3926 2/2 0:07 0:02 1:36 (100% | +1)
 
14834
Martin_Stahl
USAuPzlBxBob wrote:

Yes… we're heavily penalized for incorrect moves.

Interesting, notmtwain's recent activity.

Clearly he's stuck at the 1200 level, and primarily because he is trying to get full credit by making his moves quickly.  I've been in that syndrome myself, around the 1700 — 1900 level, and then I decided I was going to just settle for 70% credit. (the take too long default credit level)

...

 

@notmtwain posted the OP's stats..

 

USAuPzlBxBob

Oh, pardon me.  Didn't know.  Makes sense.

How do you insert the image the way notmtwain did?

Martin_Stahl
USAuPzlBxBob wrote:

Oh, pardon me.  Didn't know.  Makes sense.

How do you insert the image the way notmtwain did?

 

Using the website.

felpo4

@notmtwain Thanks, it make sense, strangely this comes suddenly after a small but continuous improve happy.png But yes maybe is kind of a change in the way I use.

marqumax

I think that the puzzles are too easy for the rating. like I'm on 2000 on puzzles but 1350 in blitz

Martin_Stahl
marqumax wrote:

I think that the puzzles are too easy for the rating. like I'm on 2000 on puzzles but 1350 in blitz

 

It's easier to find a solution when you know there is one there, than it is to recognize when such a position may be in a game, calculate if it is, and then accurately play the variations.

llamonade2
Martin_Stahl wrote:
marqumax wrote:

I think that the puzzles are too easy for the rating. like I'm on 2000 on puzzles but 1350 in blitz

 

It's easier to find a solution when you know there is one there, than it is to recognize when such a position may be in a game, calculate if it is, and then accurately play the variations.

Sure, but the scale and its placement is completely arbitrary. You could set it so beginner puzzles are -400 and master level puzzles are 20 (Elo and Glicko formulas work just fine in negative numbers).

So chess.com could try to make the puzzle rating align with the live games, just for the aesthetic value (and for not confusing new players), but I guess it's too much work for essentially nothing.

llamonade2

Or maybe not much work, but basically what they'd do is subtract 400 from everyone's puzzle rating overnight, which would confuse and upset maybe 1% of people, and on a website of millions that's a lot.

Martin_Stahl
llamonade2 wrote:
Martin_Stahl wrote:
marqumax wrote:

I think that the puzzles are too easy for the rating. like I'm on 2000 on puzzles but 1350 in blitz

 

It's easier to find a solution when you know there is one there, than it is to recognize when such a position may be in a game, calculate if it is, and then accurately play the variations.

Sure, but the scale and its placement is completely arbitrary. You could set it so beginner puzzles are -400 and master level puzzles are 20 (Elo and Glicko formulas work just fine in negative numbers).

So chess.com could try to make the puzzle rating align with the live games, just for the aesthetic value (and for not confusing new players), but I guess it's too much work for essentially nothing.

 

Puzzles get ratings just just like players do, by losing against players with ratings (being solved) or defeating them (being failed). There are some variables in the time taken to solve and problems with partial credit but in general, the rating of a well-formed tactic will get to a stable place once enough players attempt it.

 

The big swings come in the default ratings of players, but I believe the site uses RD there too, so a really low rated player shouldn't impact the stability of a puzzle's rating, if their rating isn't very stable.

llamonade2

Yeah, they'd have to decrease the puzzle ratings all at once too.

I remember they did this with rapid ratings... wow what is it now, 5 years ago? Maybe longer? Except they gave people a boost (rapid ratings were far below other live ratings at the time).

lfPatriotGames
bobgladstone wrote:

I get annoyed when I can get five puzzles correct and gain 10-15 points then get one wrong and lose 12 points

 

That's why the rating system is pretty fair. That is exactly how a game of chess works. You can get 5 moves in a row correct, but if the sixth one is wrong enough, you lose the entire game.  Just like the game of chess, you get little reward for many small successes, but a huge penalty for one bad move.

eric0022
bobgladstone wrote:

I get annoyed when I can get five puzzles correct and gain 10-15 points then get one wrong and lose 12 points

 

 

It does happen.

 

I pity FlashyFerrari for losing 114 rating points even though he obtains a partial score of 44% (from the photo in post #3.

USAuPzlBxBob

 

Annoys me BIG TIME.

Got a new highest overall score about an hour ago.  Thought I'd improve on it a little bit more, and 4 tricky little endgame studies came up, all in a row.  These are pawn races and how to defeat them.

Two of them were so impossible, and one of them, I tried every single move I could think of, to the point of:  What possible move haven't I tried?

Had a 2273 score an hour ago, lost 8 points, got the next puzzle correct for 5 back… and then lost the 4 in a row for -48.

Just dropped out of the top 30,000 Leaderboard.

Very discouraged!

cry.png

 

lfPatriotGames
USAuPzlBxBob wrote:

 

Annoys me BIG TIME.

Got a new highest overall score about an hour ago.  Thought I'd improve on it a little bit more, and 4 tricky little endgame studies came up, all in a row.  These are pawn races and how to defeat them.

Two of them were so impossible, and one of them, I tried every single move I could think of, to the point of:  What possible move haven't I tried?

Had a 2273 score an hour ago, lost 8 points, got the next puzzle correct for 5 back… and then lost the 4 in a row for -48.

Just dropped out of the top 30,000 Leaderboard.

Very discouraged!

 

 

If I could offer some unsolicited advice, it would be that you are focusing on rating too much. I had a very difficult time reaching 1800, because for me that was a threshold I didn't think I could reach. Once I reached it, I stopped worrying about rating. Sometimes I dont even do 5 puzzles per day. I spend about 20 minutes or so on puzzles per day, often less. So if a couple puzzles take 5 minutes, by the time I figure out why the solution is the way it is all my time is used up. Whether I get a puzzle right or wrong, I will usually spend some time afterwards looking at it some more. 

For whatever reason 2500 rated puzzles "feel" easier than 2000 rated ones. The only thing I can think of is ignoring the rating and time helps.

USAuPzlBxBob

 

Working diligently on improving my overall puzzle score this past holiday weekend.  Had been marching up, finally, through a recent highest score, and at 2295, I decided to keep going — try for 2300 — because it's been a goal of mine for the last month, and today I was on a roll.

Next puzzle comes up, around 10:00 AM this morning… my big chance… got the first move.

… But I just wasn't seeing it after that move… and so, like 1 hour, 20 minutes+ on it, I clicked my second move:

Wrong!  Okay, only -3 points, not too bad, let me try another move I had considered.

Wrong!  Huh!?  … another… Wrong! … and another… Wrong!

What is this move?!!!!

… another… Wrong!

I think I then tried a few I had already tried because I just couldn't find it.

Finally found the correct move.

No way, I thought!  That move?  Oh, come on!  You gotta be kidding me.

I looked at the stats for the puzzle… % Pass Rate (these are usually around 35 - 45 % Success)

This puzzle… 1.6% Pass Rate.  Less than two people in one hundred get it right!

It is rated at 2600!!!

Well, I only lost 3 points, I'll try another… I'm at 2292, the next one can't be that bad, so I try another…

I'm still working on it… I just checked its separate file window… and I've been on it for 215 minutes plus now.  Over 3 and 1/2 hours.  It is pure torture and I'm still on the first move.  It's one of those puzzles where all the pieces are in the middle, all interrelated to each other, and only a computer could get it right, probably in a nano-second.  I've had to close my eyes, take short naps, several times…, but I'm still in the hunt, at least… 

P.S.  OMG!  I just got the first move, looked at the clock and it read 300 minutes!!!!   grin.png  happy.png
Two moves later… got them correct, a measly 2+ points for a 2078 rated puzzle that took me five hours, 8 minutes.  Ha!  Ha!  (and I'm still below my highest score, and also my 2300 month long goal)
Oh well, it's been fun.
P.S.S.  Done for tonight;  tried one more, took 67 minutes:  my new rating is 2301!
happy.png

 

Poldi_der_Drache

To be honest I believe that the puzzle rating means very little, there are bad players with a good rating and vice versa. The trick is to just take more time if you are not sure of the answer and you will gain.

Also I suspect the top players get "outside assistance", there is no way you can solve these 3000 elo puzzles in a few seconds consistently. wink.png

JamesAgadir
Poldi_der_Drache a écrit :

To be honest I believe that the puzzle rating means very little, there are bad players with a good rating and vice versa. The trick is to just take more time if you are not sure of the answer and you will gain.

Also I suspect the top players get "outside assistance", there is no way you can solve these 3000 elo puzzles in a few seconds consistently.

There a limited number of high rated puzzles and so they simply memorize them.