Saturday, August 26, 2006

calculating percentage success @ CTS, note from/to spacecowboy





The following is in response to a note from spacecowboy at chessChat.org: David, I was glad to learn from your blog that your head is feeling better.

Have you noticed what I have noticed? I have recently noticed that my percent accuracy appears like it is being calculated differently on CTS than it has been calculated in the past. I have already climbed from 88.8% to 88.9% in what seems like less than one week! This would have been nearly impossible using the previous method of calulating percent accuracy. The previous method of calulating percent accuracy was to average the entire history of problems that the tactician has done on the CTS server. I suspect there is now a new CTS method of calculating the percent accuracy of each tactician which does not use the entire history of problems that the tactician has done on the server.

P. (spacecowboy)




(at left) a deranged chess dog, attempting to portray an even more deranged bobby fischer but without quite nearly the same volts, watts, joules, ergs, omms, or stanford-binets iq test juice, plots his next disparaging remark, immediately subsequent to any dktransform post at CTS, thus attempting to have others 'feel his pain as their own' in utter futility

my reply: dear p., spacecowboy:

i cannot take any big titles at CTS, but one title i qualify for is maniacal tracking in morbid detail percentage success. since i lay out an array of circles, like you get in a school exam, each night, and fill in each problem solved with a 'circle' and each failed with an 'X', i know quite well my success, in sets of eight, then sixteen, twenty-four, thirty-two, then forty, then i relog in.

if i am at 6f/44s=50tries, (or 6f/50 as wormstor prefers to represent at our respective blogs), then i will try extra hard to get zero wrong to go 6f/54s=60, and hit my haloed ninety percent, or more.



when i am at 1512/13 as i was last night, i exchange, in a way, rating for more care, and so slow a tiny bit, finishing 1510.9. my algorythm for some time, is that as long as im above 1500, preferably above 1510 for some contingency, then i try to get as many correct as possible, exactly as you do so well, and thus push up my averages. as you know so well in your remarkably logical and heuristically individual way, this approaches real chess.

1510 or 1540 does not matter to me, as long as i maintain a level that is respectable, for my skill, relative to myself. what matters to me is sustained accurate thinking, as you do better than 99% of all users there, except maybe my nemesis who we won't name, or trallala, and other names well known to you.



if someone is a blitz specialist, as i am not, then 65 to 70% is ok. but if you play as i do 3/8 (3min/8 sec increment =8:20 for 40 moves, or 3/12 as i did for many years= 11 min for 40 moves), then such a pace as 85 to 92% is a more valid test. you might sit on your hands at times, surely not at every move, as you and i do at CTS, and find the right move, even if it takes 28 or 33 or even 43 seconds, without feeling silly. believe me, when i play, i see this care show up, and know when to whip out the moves, and when to pause.

i am not sure of what you say about a new method. im more than willing to test it. if youve read any of my posts at blogger, then you know ive derived my percentage, such as when i am 83.947, then when i expect 83.9501+ look for the report back = 84.0% switched over from 83.9, so know quite well there.

when and if i cross an integrer or decimal, then if shortly thereafter i cross back, again watch, so from 72 to 84%, have gotten to do this not only 130 times, but more like 250 times, since ive gone over and back several times.

if i may please ask, what is your background? you used to be at blogger, but inactive there, despite wormstar (a.k.a wormwood), and tempo all over there mousetraper, loomis (to name but a few of many fine efforts), and now even me. wonderfull stuff. seen your comments. but, of course, recognize we cannot be all things to all persons.

i hate to edit, so send this as is now, and wish you the best.

warmly, david, from the pacific northwest usa, in seattle

spacecowboy 2nd post: I appreciate your interest and I await your determination of the issue I have raised. Again, the issue is whether the method of percent accuracy calculation has changed on the CTS server. It is likely that there are few people, if any, besides yourself who are doing the work to put themselves in position to determine the answer to this question.

and in reply to my question: "if i may please ask, what is your background?"

Later please. All in good time.

P. (spacecowboy)



dk plots his next move--after a mind expanding session with his moldavian girlfriend, musician tatiana--in rebutal to an ever move vicious attack from chessDog, after consulting zurich psychotherapists, and they all concur that he has no need to concern himself with this obviously psychologically damaged little boy.

and i replied:

dear spacecowboy: thank you for the acknowledgement that i could and would resolve this question, that is to say, whether, the method of calculation at CTS has been modified.

while i detect no change, at the moment, i just hit and passed 84.00%, so until i get a little closer to 84.049%, to thus be able to watch as to when 84.051% happens and then see a rounded report to 84.1%, and in so doing see whether this next tiny milestone occurs slight more accelerated, accelerated, or significantly more rapid, i cannot know.

after many years as a private money manager, managing many millions of dollars on wall street (as what is commonly called a 'stock-broker', only much more advanced), i have a LOT of experience with moving averages. this is like the monthly government jobs report (the graphic in the nyTimes with the moving average line is way better, but this is a start), which is always posted on the first friday of every month at 8:30 est, or 5:30 am pst, and since labor or job creation is so variable, they smooth the data, and this happens a lot in stocks, where the 50 day, 20 day, and 200 day moving averages are watched VERY closely by traders. i am not a math person, but a pattern person, either as architect, trader, self taught reading and writing basic japanese, and of course our beloved chess.

this is a feedback process, where expectations shift perception, and those in turn create rapid changes in group or crowd psychology, and in turn again alters perception. like in complex adaptive systems, some large physical systems become 'strange attractors', and clusters form, and elements around those clusters into stable constructs. add society and mind, and then we have our human world.

at CTS the true percentage success has been a cumulative summ, as you know. if the change has happened, after rest in days ahead, and get myself back to 88 and 90% success, i can pretty quickly see if this is a '1000 try' MA (moving average), or 10,000, or 100, or even an exponential moving average. thanks for your question.

david, aka dktransform

11 Comments:

Blogger Temposchlucker said...

Allthough the succesrate isn't a parameter to steer my study in the right direction at the moment, I noticed the same as Space. It seems to adapt faster.

Sun Aug 27, 01:15:00 AM PDT  
Blogger transformation said...

im a few minutes, i will hit 84.0501, and if the percentage has NOT changed, then i will post 84.1 from 84.0 EXACTLY WHEN as planned. before my last tranch of tries, i was exactly 84.0471. then this all will be settled soon.

Sun Aug 27, 01:56:00 AM PDT  
Blogger transformation said...

tempo, it just jumped right where it should be. i am now 84.058%, posted 84.1 according to expected rounding:

2783f/14,674s= 17457 tries
elo 1501.2

since this jump happened about ten problems back, i missed the exact moment when i often catch 'it', but close enought within ten. i was more focused then on a 'quickie over 1500', since i must go for a run now, its late on the west coast, must be back at work in 9:47 hours, and begin my climbing/mountaineering focused vacation in three days, so crunch time...

today 16f/121s = 137 tries @ 88.32%.

since i was 10f/40s =50 @ 80.00% from some uncharacteristic early in the day CTS, and blew it over breakfast right before work, this means i was:

tonight 6f/81s = 87 tries at @ 93.1%

... so much for chess immediately after special neck massage from a good Moldovian girl, but better at pm after a day with the public at work!

Sun Aug 27, 02:26:00 AM PDT  
Blogger Temposchlucker said...

Well, then maybe my score is degrading faster than it feels like:)

Sun Aug 27, 02:41:00 AM PDT  
Blogger wormstar said...

I too had wondered about a sudden rise in percentage, before I saw your posts about it. I had suddenly passed a couple of users in the ranking.

now it looks like there wasn't anything going on. - unless, the average has changed from infinite memory to around your number of tries. that way, you'd notice no difference, but everyone with considerably more tries would.

Sun Aug 27, 07:39:00 AM PDT  
Blogger transformation said...

worm, out of respect to tempo, you, and space who have 61k, 55k, and even 26k tries respectively, it NEVER occured to me that those possibilty exists that the calculation is for the last TRAILING 25,000 tries, or 20,000 tries.

i kind of doubt it, but how else could it be explained that temp and you both experience this alike WHILE i, who is obviously carefull with this calulation, has zero effect at 17k? this would bolster the theory of 25k. but then how come space, who is surely not given to dreams, experience this then?

i have to leave it to all you other fine big scale 'try' minds till i get to 25,000+.

Sun Aug 27, 11:18:00 AM PDT  
Blogger wormstar said...

yes, I doubt it too, but already three people have suspected something, so maybe there could be something...

the average can also be an iterative algorithm with no explicit data stored. all that suffices is to know current average and number of tries. like in newavg = (oldavg/n + latestavg)/(n+1).

there can also be factors to change the 'memory' of the average in different ways. hell, we don't even know it the avg is arithemetic or something completely different! there are wealth of mean numbers around...

it could also be that svejoe has implemented a workaround affecting only people from, say 20 tries up. especially IF they're storing all the data from day one. if that's the case, there is bound to be resource problems at some point. and as we know, there already has, and some features have been removed even before graphs disappeared...

so, we don't really know what's going on inside the code. but I guess we'll see sooner or later. I've been doing crappy percentages again, so I can't really say anything about it yet...

Sun Aug 27, 04:41:00 PM PDT  
Blogger transformation said...

fyi, i just went 84.0 from 84.1, and now am back to 84.1, so i know to the nearest try exactly where i am on CTS:

2786f/14682s= 17,468 @ 84.0584%, or 84.1% rounded, and thus i can confirm any calculations on percentage success below <25,000 tries.

Mon Aug 28, 12:51:00 PM PDT  
Blogger transformation said...

today a global roundup:

early today--
3/16=19 @ 84.2%
tonight--
1/23=24 @ 95.8%
sum--
4/39=43 @ 90.7%

2787f/14,713s=17500 @ 84.0743
1506.7 elo

once my vacation begins, ill have energy for 150 per day most days, im guessing.

it hit me that i should climb seven fingered jack, or up to 9100' of near vertical looking loose rock hard scree with a bit of ice and snow at the top. up early, all day, gain of only 5540' or about a mile... or 1688 meters vertical...

this is a towering precipice, at the end of one of those 25 mile long dirt roads that no one is driving down on a weekday in late in early september, and approaches nearby mount maude, if the fancy strikes me, and not a place to get lost on...

Tue Aug 29, 01:02:00 AM PDT  
Blogger Mousetrapper said...

I used to be at 80% and now have sessions of 90% on average. I have done roughly 25k in total. My percentage rises roughly 0.1 in every 200 tries, and this is exactly what will happen if the percentage is calculated for all problems done since start. But I admit I never did exact calculations.

Tue Aug 29, 06:15:00 AM PDT  
Blogger SamuraiPawn said...

Thank you for the warm welcoming! Always nice with a friendly voice when all I hear these days are harsh voices saying things like "Noooooo, it's Queen takes on e3 you buffon!!!" Maybe I'm going crazy... ;)

Wed Aug 30, 05:12:00 AM PDT  

Post a Comment

<< Home