Advancing Chess Tactical Server!
It seems like news has been everywhere: Tempo's 1600+ and Tempo's 1616 most impressive record high well after the notable 50,000, Spacecowboys 1474 after his 20,000, Wormwoods recent freshly minted 1533 (yowww!)after vigorously piercing 50,000 also, lots of interesting chat on the CTS message board about burning lots of problems into memory, and Markusgoths recent 25,000. It really is a daily work crowd deep in the night, like coal miners with lanterns on their head, digging deep into the earth, or in this case: chess, chess tactics, chess improvement, then lastly getting on line and beating somebody good who expects to take easy points for our rating, but we just did, well, we just did like 5,000 problems since our last game? Sound familiar to anyone??
Then my particular crowd paralleling much of the above: the etheric high percent crowd among those over 15,000 problems who are rated 1500 or above, ranked by tries: Tempo again 80.7%, nabla (FM) 87.2%, Trallala 92.9%, kawala 83.8%, spacecowboy again 88.7%, morkovkin 89.1%, chessdog in the honorably mentioned (<1500) absolutely amazing 95.7%, dktransform 83.8%, and alvis honorably mentioned (<15,000) at 88.3%.
I ask you chess.emrald.net users to go to the tactician tables and sort the top 51 active users by 'tries' and visually scan, asking yourself how many large users have done much over 83%? Nine persons.
I surveyed this by counting the number of users by percentage success only for sets of persons 60 to 61%, 62 to 63%, all the way up to 98%. I made sets of sets of those (I was VERY curious), then supersets, to simplify the core observation, and found the following (data from ten days ago):
60 to 63%= 137 users
64 to 67%= 154 users
68 to 71%= 157 users
72 to 75%= 178 users
76 to 79%= 146 users
80 to 83%= 127 users
84 to 87%= 098 users
88 to 91%= 041 users
92 to 95%= 024 users
96 to 98%= 004 users
If we look closely at chasing crowd at the "Hillary Step" (at Mount Everest in late May, their can be literally a log jamb of climbers waiting to pass at 200' below the summit, since their is so little room to pass) at the clump or cluster between 82 to 86% (85.0% is absolute my goal for 25,000), we find something quite interesting:
82%= 30
83%= 33
84%= 41*!*
85%= 23
86%= 18, etc.
In nature or complex adaptive systems, we often find other examples of 'buffering': sea tempuratures that make sudden jumbs, buffering between alkalai or acid states (I am not a chemist so please forgive errors of exact concept on this one). Why this here?
This takes me to my subject: I decided at 15,442 or 83.577% to go for 85.0 by 25k, and reversed engineered that I needed 87.183% for 9558 problems= 25,000.
Now the facts, making spacecowboys and trallalas and even chessdogs fine accomplishments all the more poingent: I had gotten used to putting it my mind: "Ok, every sixth one I can err or have errors on". 1:4=75%, 1:5=80%, 1:6=83.33%, 1:7= (natural sequence of 14, 28, 56 then repeats: 0.142856) 85.7144% or 85.72%, 1:8=87.5%, 1:9=88.9%, 1:10=90%, 1:11=90.9%, 1:12=91.33%, 1:15=93.4%, 1:20=95%.
It is very, very hard to avoid missing only one problem in seven, harder at eight. We all get those 1:10 runs where we get one wrong out of 20, then two in a row wrong, to get 3failed, 27 success=30 total. Then the bad days...
Since this decision, I have done:
12/126=138*(91.3%) Wednesday
21/147=168 (87.5%) Thursday
15/113=128 (88.3%) Friday
02/042=44 (95.4%) Saturday [1500.3 ... I tend to speed up when and if I go <15,>1520.
[or how wormwood likes to express it succinctly:
12f/138
1500.3 or:
12f/126s=138 total@ 1500]
It seems that in this clumping at 84% that is outlying data from the smooth sequence up from 60 to 79%, then down to 85%, we hit this natural resistence at 85%. And the rarety of it makes the desire of some of us % success folks who do not focus mainly on rating but more on accuracy (I am NOT suggesting that this is worth MORE, only to note this difference), makes our desire all that much stronger, since so few get there. Anyone who wishes to can try.
Some days the brain is tired, the eyes weak, the board vision dim. So we must really hit 88 to 90% on good days for a hundred problems or more since the bad days come to, such as last week when I had a terrable day at 80.5%. One mans humble opinion, please.
Then my particular crowd paralleling much of the above: the etheric high percent crowd among those over 15,000 problems who are rated 1500 or above, ranked by tries: Tempo again 80.7%, nabla (FM) 87.2%, Trallala 92.9%, kawala 83.8%, spacecowboy again 88.7%, morkovkin 89.1%, chessdog in the honorably mentioned (<1500) absolutely amazing 95.7%, dktransform 83.8%, and alvis honorably mentioned (<15,000) at 88.3%.
I ask you chess.emrald.net users to go to the tactician tables and sort the top 51 active users by 'tries' and visually scan, asking yourself how many large users have done much over 83%? Nine persons.
I surveyed this by counting the number of users by percentage success only for sets of persons 60 to 61%, 62 to 63%, all the way up to 98%. I made sets of sets of those (I was VERY curious), then supersets, to simplify the core observation, and found the following (data from ten days ago):
60 to 63%= 137 users
64 to 67%= 154 users
68 to 71%= 157 users
72 to 75%= 178 users
76 to 79%= 146 users
80 to 83%= 127 users
84 to 87%= 098 users
88 to 91%= 041 users
92 to 95%= 024 users
96 to 98%= 004 users
If we look closely at chasing crowd at the "Hillary Step" (at Mount Everest in late May, their can be literally a log jamb of climbers waiting to pass at 200' below the summit, since their is so little room to pass) at the clump or cluster between 82 to 86% (85.0% is absolute my goal for 25,000), we find something quite interesting:
82%= 30
83%= 33
84%= 41*!*
85%= 23
86%= 18, etc.
In nature or complex adaptive systems, we often find other examples of 'buffering': sea tempuratures that make sudden jumbs, buffering between alkalai or acid states (I am not a chemist so please forgive errors of exact concept on this one). Why this here?
This takes me to my subject: I decided at 15,442 or 83.577% to go for 85.0 by 25k, and reversed engineered that I needed 87.183% for 9558 problems= 25,000.
Now the facts, making spacecowboys and trallalas and even chessdogs fine accomplishments all the more poingent: I had gotten used to putting it my mind: "Ok, every sixth one I can err or have errors on". 1:4=75%, 1:5=80%, 1:6=83.33%, 1:7= (natural sequence of 14, 28, 56 then repeats: 0.142856) 85.7144% or 85.72%, 1:8=87.5%, 1:9=88.9%, 1:10=90%, 1:11=90.9%, 1:12=91.33%, 1:15=93.4%, 1:20=95%.
It is very, very hard to avoid missing only one problem in seven, harder at eight. We all get those 1:10 runs where we get one wrong out of 20, then two in a row wrong, to get 3failed, 27 success=30 total. Then the bad days...
Since this decision, I have done:
12/126=138*(91.3%) Wednesday
21/147=168 (87.5%) Thursday
15/113=128 (88.3%) Friday
02/042=44 (95.4%) Saturday [1500.3 ... I tend to speed up when and if I go <15,>1520.
[or how wormwood likes to express it succinctly:
12f/138
1500.3 or:
12f/126s=138 total@ 1500]
It seems that in this clumping at 84% that is outlying data from the smooth sequence up from 60 to 79%, then down to 85%, we hit this natural resistence at 85%. And the rarety of it makes the desire of some of us % success folks who do not focus mainly on rating but more on accuracy (I am NOT suggesting that this is worth MORE, only to note this difference), makes our desire all that much stronger, since so few get there. Anyone who wishes to can try.
Some days the brain is tired, the eyes weak, the board vision dim. So we must really hit 88 to 90% on good days for a hundred problems or more since the bad days come to, such as last week when I had a terrable day at 80.5%. One mans humble opinion, please.
posted by transformation at 3:10 AM
11 Comments:
the 84% might be because many aim for 85%, and once they get there they relax or shift more attention to rating, resulting in % decrease. same thing with even hunreds in rating.
I've noticed that there are almost always more people in 1500-1509 (33 now) than 1490-1499 (16) or 1510-1519 (23). slightly different than the thing with 85%, but I think that's because people regard percentage as a 'necessary evil' and would preferably do high ratings if they didn't believe % will yield more progress eventually. just like you know some vegetables are good for you, and you eat them even if you'd preferably be eating exclusively pizza were there no consequences. :)
At the moment, my first goal is percentage on CTS. Today I solved 96 and failed 4, losing 13 rating points.
I think it is not surprising that the distribution in percentage correct has two peaks. As you note, there are essentially two groups of people: one focused on ratings and one focused on percent correct. These two groups will have a peak in percent correct at different places and the total distribution, which is a sum of these two distributions, will have a peak at both places. I'm not sure there's anything magical about 84 or 85%.
Incidentally, I think it's a bit naive to believe that everyone is either focused on rating or percent correct. I myself notice that my rating and percent correct often both go up or both go down. This makes sense as only getting problems correct can make your rating go up.
For me personally, I am not concerned with % correct. I value the problems that I get wrong as the ones I am actually learning from. At the level of problems I am seeing, the ones that trip me up are those in which more than one tactical element exists and we have to see how they work together. I will get a problem wrong when I base my move on less than all of the tactical elements present. This is different than just whipping out a move trying to be fast. I value the speed aspect as it has been shown beneficial to do many many tactics problems in the 5-10 second range, but I still only play moves I believe in.
(i). thank you loomis. much appreciated. i must go to work in a few minutes, so cannot respond (for now please) more than very briefly.
i dont think that i said that EVERYONE was either focused on rating or percentage so to say its bit niave sounds a little bit harsh, but i get your point.
i also work on both, but, as i said, value the latter a bit more as for me it translates better into much more accurate OBP. if i was platued at 1479-87 etc, as i was at most of june, i might feel differently. similarly tempo as he was previously in 1540 (or so), then latter high 15's.
there is a lot of talk about ratings at CTS, much of it very good. i only wished to analytically place the % into further resolution, wrote a brief comment on the high problems solved and four star users THEN wrote about percentage and my own experience of it--AGAIN--never saying that there were two groups.
(ii). as for my OBP, i think that with the proviso that if an individual has already done a LOT of problems (>10 or 15,000), while they are not yet at the 'super CTS star crowd' >25,000 (i.e. term 'super GM'), such a person focused on, say, getting every 1/7=8 (0.875) or 3/22=0.88% or 1/9=10 (0.90%) would see a demonstrable improvement in their own rapid or blitz play.
IMHO, i think that if a person made one bad or inaccurate move in every 8 in blitz youd be killed.
it seems to me, that the marginal difference between habitual 8-14 sec problems and 3-7 is marginal for CTS yet maximal for OBP.
not as fast as the 74-78% crowd, but fast enough for most blitz still. it is not my intent to evangelize or dogmatize two groups, only to pin point what i have noticed. thank you.
lastly, there are many games to play, and again, i only wish to remind that their are other games played here by some. of course, some play many games. for example, right now, i am highly motivated to A. get to 20,000 soon while also preserving accuracy. and B. i am highly motivated to get in the top 40 tries crowd, with me at #43.
Did you notice the bellcurve of succesrate http://chess.emrald.net/ctsTactHome.php
For me (as for CTS) there is no real difference between a problem that takes me a lot of time or one that I do wrong. Both I have to work on to commit them to LTM. Most problems are so simple that anybody would do them right if they didn't mind about time.
(iii) it seems to me that the marginal difference between habitual 3-7 sec CTS users and 8-14 users who sometimes use only 4 seconds, but when presented at times with a thorny problem can go 20, 30, 40, even 50 seconds for OBP, see maximal difference in rapid or blitz results, since one accurate 'deep think' on a tough move on the clock at times can be the difference between being killed tactically, verus a single quiet move or strong, imaginative move which wins.
lastly, again, i find that to run 1:8 or 1:10 WHILE watching rating and the clock to be very, very hard, my main point. we get tired, we misconceive, we fail to see our opponents move, in short, we fail--so to sustain this sharpness is exhausting. ask wormstar! as chessdog at cts.
but you already know this, since you also suggest your focus on accuracy.
i think that to do 100 or 120 problems carefully is no less than 160 or 200 more quickly. we must go back, we must pause, we must retry failed. lots of stuff. not that 77% users dont, but its a lot more finicky. blessings, time to go. thx.
DK,
I'm sorry, I didn't mean to imply that you had said everyone must focus either on percentage or rating, just that there are people who seem to focus on one rather than the other, or value one over the other. Following that, it was me who was naive (and this is what I meant in my second paragraph) when I said the total distribution must be the sum of two separte distributions with distinct peaks. I do think this is the correct reason for the two peaks in the distribution (rather than something particular about 85%), just that the explanation as I gave it was very coarse (by naive, I simply meant a lack of sophistication in the explanation).
I hope that helps to clarify what I meant.
Secondly, I agree that focusing on getting problems correct is a good way to study (as is doing lots of problems rapidly). This is, after all, a closer approximation to playing in a game than doing problems quickly.
As you note, there is a lot of interesting talk about rating and % correct on CTS. I think I should organize my thoughts on these and perhaps will soon have a coherent entry on my own blog.
It's not clear that the higher value at 84% is interesting. It could just be a statistically unsurprising fluctuation in a noisy data set. Unfortunately, that is a hard problem to address quantitatively (i.e., to put a p-value on) without taking lots of time to think about the best way to do the analysis.
loomis, i am very appreciate of you and what you say. i greatly look forward to further dialogue with you directly (both here and at your very smart post). you way of thinking and expressing is excellent.
im home for lunch, from work, on the west coast, for my five or ten minute nap, so must run NOW. seven days at work in a row (required in my case this part of the rotating month), then time off.
this will allow me to run late tonight after tactics and ctArt and a few slow blitz (3 min/8 sec increm), so i can go one mile or 1600 meters up in 3.5 weeks, quickly and relatively painlessly, even though i walk eight miles at my store per day, i must do this....
blueDevil, i have missed you a lot, and need to get up to date on your wonderfull posts. i appreciate all you say, also all very smart. im a very, very visual person, so despite being an architect/financial/systems guy, not formally trained in math or science or computing, so much great stimulation for me to visualize and reflect on.
thank you all. i must sleep now for ten minutes! great joy!
Trans,
I heve not worked with CTS so I can't say. I have some prejudices that do not bear much weight given that all I know of CTS is from reading blogs. My preference is to work with a much smaller sample and not put myself under such time constraits. To do this I have used the Bain book and ct-art as well as flash cards I made up from one of the Blok books. My belief unfettered by the constraints of having to prove it 8) is that smaller samples sets with shorter repetition periods leads to quicker results.
Post a Comment
<< Home