blitz rating system suggestions

[uri blass]

1)In CTS some seconds are given for free and there is no difference in rating between the case that you solve the problem in 1 seconds and you solve the problem in 2.9 seconds.

I think that there should be something similiar in chess tempo.

In this site I never can feel confident of not losing rating points even if I solve the problem fast because my speed is compared to other users who solved the problem correctly.

There were clearly cases when I lost blitz rating inspite of solving the problem correct in less than 20 seconds and if I understand correctly in theory even if I solve the problem in 2 seconds I may lose rating points if everyone except me solved the problem in 1 second.

2)The time that people need is not considered in calculating rating of problems and I think that is not fair.

A problem can get low rating when even for low rated players it is easy to guess the solution
inspite of the fact that everyone including high rated players need time to understand why the solution is correct.

These low rated problems can cause high rated players to lose rating even if they solve the problem slightly faster than average.

Uri

[tama]

That's why i don't practice in  Blitz mode, not enough time to fully calculate.

[richard]

You do get 5 seconds to look at the problem without the clock ticking, this is however before the opponent's move which isn't exactly what you want.  Yes your speed is compared to other's. Essentially CTS isn't much different, the time allowed to solve a problem before you lose points is I believe (from memory) related to the rating of the problem, which is in turn related to others performance on the problem.

I agree with you that there is an argument that the time taken should also effect the problem rating as well as the user rating, but I don't believe the number of problems where this would make a difference is great enough to introduce a change in the rating system at the moment, if I have to make changes for other reasons in the future I may consider making this change at the same time (changing the rating system is a big deal as it essentially invalidates all previous problem and user ratings).

Regards,
Richard.

[uri blass]

 
In CTS The time allowed to solve a problem before you do not get the maximal bonus is exactly 3 seconds.

It means that there is no difference in rating if you solve the problem in 3 seconds or in 2 seconds but there is some difference in rating if you solve the problem in 4 seconds relative to 3 seconds.

based on my understanding it is not the same in the blitz rating in CT 
Maybe there is a problem to change it because it is going to make old rating invalid when you have new conditions but I think that it is better to give the user some free time so he can be sure that solving the position correctly and fast is going to give him rating points(because based on my understanding in theory people can lose blitz rating points for solving a position in 3 seconds if all the other solvers solved the same position in 1 second).

Uri

[drahacikfm]

The theory is nice, but have you ever seen a problem on CT with an average solving time less than 3 seconds?  I have never seen one less than 8 seconds.  Maybe there are a few, but not enough to even discuss the issue.  Looking at the last 50 problems you did in your Stats, the shortest average solving time is 17 seconds, and most of them are more than 30 seconds!

If you read the FAQ you will see how the blitz ratings work.  There is a certain amount of "free time" for each problem, which varies from one problem to the next.  Rating changes are based on the average solving time and one standard deviation faster.

For example, if a problem has an average solving time of 15 seconds, and one standard deviation faster is a solving time of 10 seconds, then if you solve the problem in any time from 0 seconds up to 10 seconds you get exactly the same rating change.  So you have 10 seconds of "free time", much MORE than on CTS.  You will never have less free time than you have on CTS, because no problems here on CT have average solving times less than 3 seconds.

So this "problem" is not a problem and there's no need to change the rating system for something that might happen in theory, but which never happens in reality.

[uri blass]

I read the FAQ and there are only general rules and not exact explanation how the rating works

From the FAQ:

"For blitz style games, time taken to solve problems is considered and an extra bonus is given for problems solved quickly. The solver gets a larger rating increase for problems solved quicker than one standard deviation from the average time taken for that problem."

It is not clear how much larger rating increase
and it is not clear if the bonus is the same for
1.2  standard deviations faster or 1.3 standard deviations faster.

Note that the standard deviation may be bigger than the average so it is possible that the solver never get a larger rating increase even in case of solving in 0 seconds.
 
I will give a simple example

Suppose that 9 people needed 1 minute to solve the problem  when 1 person
needed 11 minutes.

average time per solution is (9*1+1*11)/10=2 minutes

variance is (9*(2-1)^2+1*(2-11)^2)/10=(9*1+1*81)/10=9

standard deviation is 3 minutes.

In less extreme example the standatrd deviation may be big enough so only people who solve the problem in less than 2 seconds get a special bonus
so there can clearly be a difference in rating points between solving the problem in 2 seconds and solving the problem in 3 seconds.

Uri

[drahacikfm]

It is not clear how much larger rating increase
and it is not clear if the bonus is the same for
1.2  standard deviations faster or 1.3 standard deviations faster.

This was discussed in the forum a couple months ago and Richard explained it exactly.  Lets say a problem has average 15 seconds and one standard deviation less is 10 seconds.  Then any solution between 0 and 10 seconds gets the fixed bonus.  Doesn't matter if it was 1 second or 9 seconds.  Any solution between 10 and 15 seconds gets the normal rating change.  Again, doesn't matter how much time was used, as long as it was in that range. Solutions more than 15 seconds get less and less rating as the time increases.  At some point (not told to us) the rating change becomes negative for a correct solution.  The most you can lose on a problem for a correct solution is if you take infinite time, and in that case you lose the same amount as if you got the problem wrong.

I think Richard modifies the deviation so that it is always less than the average time.  He will need to verify that.  So it is always possible to get a bonus, and it is almost always, if not always, more than 3 seconds solving time.

And it's obvious to me that the rating system on CT is better than on CTS.  On CTS, if there is some very difficult problem that has an average solving time of 60 seconds, you can only get the maximum bonus by solving it in less than 3 seconds?  That's silly.  You should get the maximum if you do it in less than 20 seconds or some other amount related to the average.  Not some silly 3 seconds for every problem, no matter how difficult.  CT's method is clearly better.

Anyway, these issues are not really that important.  Improving in chess is important and you don't do that by worrying if you took 2 seconds or 3 seconds to solve a problem.

[uri blass]

I agree that the rating method in CTS is not very good but I do not think that the rating in CT is better.

The method in CT can be better if some changes are done.

1)Average is effected by extreme value and one user who thought for an hour can cause the average to be 3 minutes instead of 1 minute.

It is better to use median and not average.

2)using standard deviation is also not a good idea because the standard deviation is also effected by extreme values.

It is better simply to give bonus for people who solve the problem faster than 75% of the other people.

3)It is better to have a rule that everybody that solves the problem in less than some default time get the maximal bonus and the default time can be 1 minute.

Maybe it is possible to have new rating list when time is important like OTB games but not blitz because in blitz I remember cases when I lost rating for solving a problem in less than 20 seconds and unlike CTS when I try to solve problems faster I usually get a wrong answer when I imagine that the problem is easy when practically there is some trap or some simple capture that I did not consider(maybe part of the problem is that I see the board less time before getting the first move when in CT it is 6 seconds and here it is 5 seconds).

In CTS there were few cases when I got the solution in less than 1 second.
I am not sure if it is possible to get here the solution right in 1 second even if you know the problem but maybe I am wrong and maybe it is also dependent on the computer that you use.

I may try to check it if I can train unrated and try to pick a move as fast as possible to see if my time is less than 1 second. 

Note that there are cases in CT when I see the clock continues to run even after submitting the last move of the solution.

Uri

[drahacikfm]

I agree with your #1 and #2.  Those could be improvements for CT.  But even without those, I think the rating system on CT is better, as explained in previous posts.

[uri blass]

another thought about it:
Time of the solvers may be misleading because there are problems when weak players find the right move without thinking when strong players are more careful and think maybe there is a trap.

These problems cause strong players to lose rating points.

Here is a simple example
problem  23642
The right move is obvious but I thought more than 40 seconds to check what happen after NxQ to be sure it is winning and white has material advantage so I lost rating.

 

[uri blass]

It is not clear how much larger rating increase
and it is not clear if the bonus is the same for
1.2  standard deviations faster or 1.3 standard deviations faster.

This was discussed in the forum a couple months ago and Richard explained it exactly.  Lets say a problem has average 15 seconds and one standard deviation less is 10 seconds.  Then any solution between 0 and 10 seconds gets the fixed bonus.  Doesn't matter if it was 1 second or 9 seconds.  Any solution between 10 and 15 seconds gets the normal rating change. 

This is simply not correct.

I solved a problem faster than average and got negative change in rating

Here are the details

 Play: 00:43

Problem #: 49065

 
--------------------------------------------------------------------------------
Problem Rating: 1790.9
Average seconds: 66.2
Attempts: 49
Success rate: 59.18%

I used 43 seconds for the problem
My statistic say that 41 of them were after the first move but it is only because I tried simple fork that win after 2 seconds and had to think later about the computer move.

The details from my statistics page are slightly different and basically they say that I used 42 seconds when 40 of them are after the first move

Here are the details copied from the statistics page:

2 2008-10-10 17:09:45 49065 1792.5 Blitz - Rated 65 42 40 2183.4 (-0.7)

[drahacikfm]

This is simply not correct.

Everything is correct.  You are forgetting that time spent after the first move counts twice as much.  So if you spent 42 seconds total, and 40 seconds after the first move, it counts 42 + 40 = 82 seconds.  Slower than average of 66.  Because the problem was almost 400 rating points below you, you lost rating points because you took longer than the average.

[uri blass]

I understand inspite of the fact that part of the 40 seconds were before finding the first computer move but after playing winning move that the computer rejected.

The question is also if the average time is based on the total time that I used
or on the corrected time.

In other words is my result considered as 42 seconds or as 82 seconds for the average?

In the first case it means that people can get correct solution with above average performance and still lose rating points.

Uri

[uri blass]

I wonder if high rated players lose rating points in blitz when they play against easy problems.

Is there some statistics about it?

In my last problem I earned 1 rating points but I used significant time to see that there is no trap because I am not used to get easy problems.

I used 26 seconds when the average time is 24 seconds.

http://chesstempo.com/chess-problems/15365

The point is that weak players may get often easy problems so they do not suspect that there is some trap when they see some piece hanged and they can solve the problem fast when stronger players do not get often easy problems so they have real reason to suspect and be more careful not to hurry to capture hanged piece and to check that the opponent has nothing against it.

Edit:I posted other thoughts about rating in the following thread.
http://chesstempo.com/chess-forum/site_feedback/oh_no_more_problems-t396.0.html

I will only repeat my main point that is that it is possible that there are problems that weaker players get better result than stronger players in them for reasons that I explained in my last post of that thread so in extreme cases the rating of problems can be mainly result of the initial rating and initial high rating can remain high (because people with high rating that are most of the solvers do not solve the problem) when intiial low rating can remain low because people with low rating
more often solve the problem for the wrong reasons.

Uri

[drahacikfm]

The point is that weak players may get often easy problems so they do not suspect that there is some trap when they see some piece hanged.

I have done every problem on the site rated under 1400 Standard, more than 10,000 problems.  Yes, there are some problems on here like that, but it's certainly less than 10% of the total problems.

Changing the rating system is not the solution!

The solution is to add more moves to the problem.  Then to get the problem correct you have to show that you can handle the trap correctly.  The low-rated players will then get the problem wrong and the problem rating will go up.

This is related to problem "pruning", and there have been many good discussions in this forum about that already.  Richard said he will be working on improving the pruning in the future so there won't be so many one-move problems.