rating

lackinglatin
Posts: 1
Joined: Sat Mar 21, 2015 6:40 am

Re: rating

Post by lackinglatin »

How interesting. I just came here to express the view that the problems seem quite hard. I've been around 5kyu for years, but hardly play for the past several years, and I struggle to get higher than 12kyu. I guess my global strategy is much better than my local tactics. That would make sense, as I always have a habit of getting a large lead in the beginning, and then making seemingly obvious mistakes and witling my lead down up to the end. I just always assumed it was a lack of attention to detail; perhaps I really just need to use this site more.
iopq
Posts: 19
Joined: Sat Aug 16, 2014 1:37 am

Re: rating

Post by iopq »

lackinglatin wrote:How interesting. I just came here to express the view that the problems seem quite hard. I've been around 5kyu for years, but hardly play for the past several years, and I struggle to get higher than 12kyu. I guess my global strategy is much better than my local tactics. That would make sense, as I always have a habit of getting a large lead in the beginning, and then making seemingly obvious mistakes and witling my lead down up to the end. I just always assumed it was a lack of attention to detail; perhaps I really just need to use this site more.
two issues at play:

1. It's easy to do nothing but 15k problems and get 1 dan.
2. The rating of the problem itself

I think the problems are about right (although I solve 1 dan problems frequently), but me holding 4 dan rating is not reasonable because I'm a 3k on KGS
Maybe someone who's 4 dan needs to be able to solve 4 dan problems?
zwim
Posts: 18
Joined: Sun Sep 25, 2011 7:10 pm

Re: rating

Post by zwim »

Your rating really depends a lot of the way you are using the site.
For instance i try to do all problems that are presented to me(25% complete currently), my rating is stuck around 9-10k because of this, i win a lot of problems but also fails a lot of them and once you missed one, even if you win it later the damage to your rating is done.
On the opposite if you come fresh, do only easy problems and succeed all then your rating only goes up and you end up dan level which is wrong.

For the problem themselves, from what i can observe about the ones i created, their rating evolves largely over time.

For instance let's take this one : http://www.goproblems.com/17920?psetid=10688617
As you can see it is rated 6d now, but do you know it stayed 5k for at several months with many attempts ! After a moment it was less frequently played, a couple of well rated players failed it and the pb rating raised to 2k and maybe because of a psychological effect, people were afraid about it and it raised 1d then now 6d, but frankly this pb is not that hard.

Also, high rated pb are rarely played, because most people search for problems based on rating, so problems above 2-3k are less frequently played and pb above dan level lesser and lesser played, thus once a rating problem has diverged, there is little chance it goes down, unless a bunch of ddk fall in love with it and crush it down.

On the opposite the problems in the range 8k, like this one : http://www.goproblems.com/17789?psetid=10688617
Are pretty stable, since it was created, its rating is stable, because it is played by the majority of players.

The rating of a problem sometimes converge but not always, and some problems have a big (even huge) volatility, starting from there, players rating cannot be accurate either and so on.

My opinion is that we should have 2 systems going simultaneously.
Have players enter their rating from external server (kgs, igs, whatever...) and deduce the problems rating from these fixed players ratings.
And a second system to deduce players resolution efficiency (a score, not a kyu/dan level) from the problems ratings.
This way we could avoid the non-anchored issue happening with ratings i described above.

The inconvenience with this is, it is accurate only if people are truthfull in the rating they enters, and also they would have to update it regularly.
The external reference server is not that important, there are are comparison tables to adjust ratings, and even if debated, it's no more than 2 kyu difference between effective rating and what's given in the tables, compared to what i read in this thread, this is no big deal.

Another possibility is to erase completely the double reference between problems ratings and user ratings.
And rate problems only according to % of succesfull solves. A problem solved by 95% of players is 20k, by 5% is 2d and so on.
And allows for the same player to solve it multiple times, but retain the win/loose status only if there is sufficient delay between the tries.
The advantage of statistical rating is that it is assured to converge.
Post Reply