tomohawk vs. drahacikfm

[drahacikfm]

Tomohawk and I keep fighting for the top spot in the standard ratings!  When I see he is #1, I do a few problems to pass him.  When he sees I'm #1, he does a couple problems and passes me.

Sooner or later one of us (probably me) is going to get a couple problems wrong and drop like a rock 25 or 30 points.  It's hard to go up when you get one or two points for a correct problem, and lose 10 or 15 for a wrong problem!

[drahacikfm]

P.S.  Didn't mean to leave out xtruder.  He's sometimes there at the top too.

[revenant]

drahacikfm, that's amusing :-)  Do you find that the friendly competition is improving your ability?  It would be nice to know that your hard work is paying off somehow in over-the-board play.

Richard, are there similar contests taking place among the top-10 Blitz Ratings?

As the site starts to attract more and more serious players it will be interesting to see if the current top-10's disappear under an ocean wave of new talent.

[richard]

Hi Revenant,

I'm not sure about blitz, I've been enjoying watching the see-sawing contest in standard though :-)

Because blitz is the default rating choice it often sees very highly rated players drop by just to give the site a look, do just enough problems to become active and then move on.  The standard ratings tend to be made up of more consistent users of the site. I had a look at the the blitz top 10 users a few days ago and there were very few non-titled players in the blitz ratings, the top 10 is currently made up of 1 GM, a few IMs and FMs and only 1 or 2 non-titled players (although it is not always easy to tell, they may be titled also).

Regards,
Richard.

[tomohawk]

Sorry been on buses and planes the last day or so, thus didn`t get a chance to post before now.

my general goal on this site is two-fold

1. To increase my accuracy. i am generally not happy with anything less than 90%. I  have a long way to go to get there in blitz. I personally think that in chess, accuracy is by far the most important thing when doing analysis.

2. To improve my rating, particularly in Standard. I  am not sure how this site would compare to FIDE ratings, but my standard here and FIDE are almost identical.


I did find it amusing when xtruder, drahacikfm and I were battling for top spot. So I did a problem to move up to number one to give drahacikfm something to shoot for.    By the time he reads this it wouldn`t surprise me if I were back to number two.

[drahacikfm]

Damn!  tomohawk 2369.8  drahacikfm 2369.5.  Big 0.3 lead!

[drahacikfm]

Ok, big boy, take that!  drahacikfm 2370.5, tomohawk 2369.8

[revenant]

lol

[slacker00]

While you guys battle it out for #1-#3, tama & roger & I have been battling it out for #4-#6.  Those guys sprinted way up there, maybe I can catch them this weekend.  I see revenant is right on my tail as well.

[drahacikfm]

It's amazing that tama is in the top 6 even though he played the first tournament of his life this past week.  He must have a great tactical talent, and might become a very strong tournament player, even though he can't spell to save his life

[drahacikfm]

Hey tomohawk, that's not fair!  A 16 point lead?  You are supposed to inch up 1 or 2 points at a time, not take a huge lead like that! 

[tmr]

The race to the every increasing top shows that the rating system isn't working at the upper end. 

Richard mentioned previously that the Glicko system accounts well for the sparce problem set at the upper end but I think it can only do so much.  This isn't a big problem but it seems as if the top rating in standard mode goes up about 10 points per day on average.  Given the static problem set it seems as if the upper ratings in a truely balanced system would be more static.

While users do lose more points for getting problems wrong than they gain by getting them right, the proportion of problems missed by the top players is fairly low so the ratings on the upper end just continue to increase.  Without enough high level problems I think the rating system needs to be adjusted so that at the upper end the points lost for missing a problem rated at less than your rating essentially takes away most of the points gained by getting the last X problems less than your rating correct.  Thus the main way to increase your rating is to get problems rated above your rating correct.  This corrects for the lack of high rated problems.

As it is now the upper rating levels increasingly reflect the number of problems done.  This creates a bias against new users (even those that become high rated) that is hard to overcome since they essentially have to do as many problems as the current top rated users to reach the same level. 

Of course this isn't the case for lower rated players and most of us don't have to worry about this as we'll never really reach those high levels anyway.

Of course on the flip side I guess if enough high rated problems can't be found the only way to encourage high level players to return is to let them battle it out as is happening now.

[richard]

Hi tmr,

The rating system WILL deal with this eventually (unless I have a bug I don't know about yet - which is possible), although it is annoying how long it takes for the users to reach their "level".  Assuming tomohawk and drahacikfm keep their success rate stable they will eventually reach a point where their rating can't go higher. As their rating goes up they will be getting an increasingly small amount for getting problems correct and an increasingly large amount for getting problems wrong and this will lead to an equilibrium at some point.  Once that point is reached further improvements require either a) They get better at tactics or b) They get better at solving chess tempo problems.

Hopefully a) and b) have some overlap but increases in b) are possible without increases in a) as users over time do get used to the types of solutions the generator will expect, for example fairly early on you work out that a tactic that wins a pawn is unlikely to be correct, later on you might also realise that if you've found an exchange winning move you better make sure that there are no better moves as with the current "alternative winning" threshold the exchange might not be good enough if there is a better move.

The main issue with how glicko is operating at the moment is that it does take a fairly long time to reach equilibrium.  Drahacik did appear to have reached the equilibrium point before the alternatives update was made in June.  Obviously he hasn't quite reached a new equilibrium yet (although I suspect he may be getting close [which might be good news for tomohawk :-)]).  Drahacik is spending a lot more time these days doing custom problem sets than rated problems so it is not surprising that it has taken a while to approach equilibrium.

I can probably look at tweaking some of the glicko parameters to try and get  people moving towards equilibrium a bit more quickly (keeping RD higher for longer should help a bit for example).

Of course whether the rating system "works" here or not is not the most important thing, it would be nice if the top rated players were able to be given harder problems and until more are added to the set this is unfortunately not possible, the current generator run is approaching the end and that means I can hopefully start to generate new problems again soon.

Regards,
Richard.

[tmr]

Thanks Richard.  I guess that makes sense.  Obviously I'm not familiar with the parameters or how they influence rating convergence but it seems as if convergence for the current group is quite a ways off. 

I'm suppose though that the current settings are probably about right given that it adds a bit of excitement about getting to the top of the list.  A more static top ten list might be boring.

[drahacikfm]

I think the rating system is working.  A couple weeks ago I was over 2390.  Then I did about 5 problems wrong in a session of about 10 problems, and I dropped like a rock down below 2330, losing about 70 rating points.  I still haven't gotten back up to where I was before.

Now I am trying to be much more careful, and I am taking much longer on each problem.  Much longer than the average user's time, even though I am probably better at tactics than the average user.  I have to get about 10 problems right for each problem I do wrong, to stay even.  Recently I have done a little better than that, so it's still possible to go up.  As my rating goes higher, I might have to do 15 problems right for each problem I miss, and that might be impossible.

For users with lower ratings who score about 50% correct, it's not a big deal to miss one problem.  They can get the rating points back on the next problem.  But for people at the top, it's a disaster.  You then have to work for an hour or more and be perfect on 10 problems in a row to get those points back.