Saturday, February 7, 2015

Back to Blunder Prones Blunder training

I do tons of puzzles at Chesstempo, CTS, Chess.com, Chessimo, Software from chessok.com, Chessity.com and and and. Lately i did start to repeat the Chessimo ( Strategy ) again and i did feel sorry that there are no new lessons for Chessimo. But after some thinking i remembered a Method of Blunder Prone to make your own puzzles: Analyse your own games and make puzzles from them.
The drawback of this idea is , you get a loooot of puzzles if you put your games into an engine and put these positions in a spaced repetition software ( like Anki ).

So i use now this algorithm:

  1. Play game 
  2. Do a Blundercheck with Fritz ( and stockfish 6 ) Threshold = 0.30 Time=1 sec for my moves ( Link ).
  3. Then i start "Infinite analysis" ( Link ) and go to these blunders. I try to understand what i made wrong and why the best move of the engine is better. 
  4. If the evaluation of best move of the engine is 0.2 better than the evaluation of the second best move of the engine and i think i understand the move then i use the Windows Sniping Tool to copy the position and create a new Anki Card with the position as question at the front side of the card ( just an "insert" ) and the computer analysis at the back side of the card.( Link )

Example:
Front side:


Back side:

r1b1nr2/pp3qbk/2p3p1/3p1pPp/5B1N/1PPR1P1P/P1Q2PB1/2K1R3 b - - 0 1

1. =  (-0.21): 1...Nc7 2.Rde3 Ne6 3.Bd6 Rd8 4.Be5 f4 5.R3e2 Re8 6.Bxg7 Qxg7 7.Re5 Bd7 8.Bf1 Nf8 9.Rxe8 Rxe8 10.Rxe8 Bxe8 11.Ng2 Qe5 12.Qd2 Ne6 13.h4 c5 14.Qe1 Qxe1+ 15.Nxe1 d4 16.Bc4 Bd7 17.Nd3 dxc3 18.Ne5
2. +/=  (0.43): 1...Rg8 2.Be5 Nc7 3.f4 a5 4.Nf3 Re8 5.Rde3 Bxe5 6.Nxe5 Qg7 7.Qd2 a4 8.b4 Be6 9.Kb1 a3 10.Nf3 Bf7 11.Rxe8 Rxe8 12.Rxe8 Nxe8 13.Ne5 Be6 14.Qc1 Nd6 15.h4 Nb5 16.Bf3 d4 17.cxd4 Nxd4


I did look for 1...Nc7 but i did not see 2...Ne6! to block the e file and make pressure to g5 , f4 and d4 ..
So i have now a nice example of : Open file, centralisation, improving of the worst piece, prophylaxis..

There are 2 things to do: Learn new knowledge and apply old knowledge

Wednesday, January 28, 2015

Stockfish 6 is here

Stockfish 6 is now available for download here: https://stockfishchess.org/

Stockfish 5 was the strongest free engine and together with Komodo 8 the strogest of all chess engines. I did test Stockfish 6 vs Komodo 8 in bulletgames and at my computer Stockfish 6 was clearly stronger. But Komodo8 is not that good in quick games anyway. We will have to wait to get the new ratings.

Edit: first ratings at Ipon: http://www.inwoba.de/index.html    Stockfish did gain 38! points and is now clearly the no 1 top enegine

Edit:  http://spcc.beepworld.de/index.htm  http://spcc.beepworld.de/endless-roundrobin.htm

Edit:  http://www.computerchess.org.uk/ccrl/404/

Tuesday, January 13, 2015

Salty repetitions

About 20 000 attempts in exactly 2400 different real OTB "Mate in 1" - puzzles and little more than 10%  speed gain in unknown puzzles of that type, that is not really exciting.  If i have to repeat that with every type of tactical theme then i will need to get 3000 years old.
To gain speed, i need to repeat the stack of new puzzles every day because i solve to many "old" puzzles too. I repeat this new stack every day until my speed did increase 25% . But this method is seemingly not (much?) better than Tomasz method. Now what? Less repetitions? Of which puzzles?
More puzzles per day? Or even less?

If i want to reach 40 Mates / Min then i will have to continue for a while. Time to find a better method.

Friday, January 9, 2015

Analysing statistics at ChessTempo

Its hard to tell if you did improve in Tactics or not, or compare your performance at one type of tactics to a different one. The time you need to solve puzzles is an important factor. Chesstempo offers for premium users a statistical tool which can be used. At the users statistic page they can find a button : "Show Filters".  Here you can create subsets of the puzzles you solved and see some statistics about these subsets.

Duplicates Status : Select Non - Duplicates
( Duplicates are puzzles you have seen more or less often, they show how good your memory is and not how good you can solve new problems)

:
Problem Set : Should be Blitz or Mixed but Standard may do too

I did filter for 2 different criteria here and got a set "Current Filter" and "Previous Filter":


As you can see: both sets are without any "Duplicates". Seemingly i did perform differently at both sets because the performance rating is different ( Change = 189.3 ). The performance rating calculates a rating based on the different "Average Problem Rating" ( Change = 231 ) and on the different "Accuracy" ( Change = -5.1 ).  But my thinking time was different too. In comparison to all solver my average time was 1.76 times slower at the "Current Filter" but only 1.13 times slower at the "Previous Filter". ( You need to calculate in sec! 2:31 = 151 sec, 1:26 = 86 sec and 151 / 86 = 1.7558... )

Both relative times are bigger than 1, so my Performance rating is in both cases "to high" . But my relative time was much higher at the "Current Filter", does this explain the "big" difference in the performance rating between both sets of 189.3 ! ?

Sadly the answer is dependend on a fuzzy number we usually don't know: "k". k tells us how much our accuracy increases when we calculate longer than "all solvers".
k seems to be "usually" between 100 and 200, dependend on tactician and their form.

But we can try to calculate k from the given data and see if it is "uncommon"

So lets assume the difference in performance rating would be the result of  the difference beteween the both relative thinking times

189.3 * log(2) / ( log(1.76)-log(1.13) ) =  56,98  / 0,1924 = 296,1 = k ?

This seems to be to high

The difference between both subsets are estimatingly not only caused by the difference in thinkingtime

Of couse the calculation is simple if the relatice thinkiing time between both subset is more or less the same. Then we only need to look at the Change in the Performance Rating

Edit Jan,12

I was looking at my "progress" in tactics from 2010 to 2014 with the method from above:


The improvement in Performance rating is easily explained with : k ~180 , so the former work in the saltmines and solving round about 100 000 tactic puzzles had only small effect ( if any )

Friday, January 2, 2015

Chess and Chat

Chess Cube does have chat rooms, but they have so many of them so that most of them have so few people, that nothing happens. The administrators of these rooms are very "strict" : that does not make the people to talk more.
I found an alternative : the debate+ room at chess.com. Chess.com did change their architecture of chat rooms. Now the chat rooms are organised by privat "groups" at chess.com.
To be able to chat at debate+ you have to
  1. get member at chess.com
    its free of charge, chess.com is a site with 12 000 000 members, about 15 000++ are playing every moment in "live" online games (bullet,blitz, rapid...). Others prefer "turn based games" where you have several days for your move
  2. get member of the group debate+   
    free of charge, can be quit at any time
  3. Play some games at live chess
    was it 10? games
  4. and wait 3 days until you are allowed to chat
    This rule of chess.com suppose to prevent "Trolls"

Wednesday, December 31, 2014

Salt fewer

At the beginning of my Salt mining i solved 300 - 600 mate in 1 puzzles / day. Now i am able to solve 900 and above each day. Still its hard. I get unbelievable tired, almost knocked out. I had a related experience with the dual-n-back training which is supposedly helping to improve the intelligence. Lets hope that's a good sign :S .
My speed at the 1800 puzzle subset did increase by ~~25%. Soon i will increase the set, or replace old puzzles by new ones and try to increase the daily "salt output" to some thousand.
Then i can tell if transfer happens with this method ( too ).

Transfer is the key question: does tackling puzzle  A help me to solve the puzzle B. With my gigantic puzzle collection it is easy to check that. Tomasz did solve thousands of mate in 1 puzzles per day and had after 50 000 or 60 000 attempts a nice improvement. So improvement in speed seems to be possible, at least at this microscopic subclass of tactics. Next question is: is there a transfer to tactics as a whole? Or maybe even to OTB? And: how to improve on "not mate" problems...
The nice thing about the universe.. there is still a lot of time left.. 2015 will be here in about 1 hour.

Happy new year, dear reader.


Sunday, December 21, 2014

In the salt mines

Maybe it was Dan Heisman who did "invent" the salt mines. He suggested to learn to solve easy tactic puzzles at hyper speed. Easy puzzles are the building block for harder puzzles. A mate in 7 is one move plus a mate in 6 is 2 moves and a mate in 5....
If you cant spot a mate in one easy then you will never find a mate in 3. If you are extremely familiar with mate positions you will see the final pattern far ahead even without calculation.

Heisman compared the solving of a tactic puzzle with making a ( complex ) multiplication. It makes no sense to learn the result of 15472894 * 94939002 by heart, but it makes sense to learn the standard multiplication table ( 1..10 or maybe even a little higher ) well and then learn how to use this knowledge to solve more complex tasks.

So Temposchlucker learned to solve sets of easy tactics at high speed and he did call these exercises "Salt mine".

Some years ago i did mine some salt too. I did create a set with 4000+ easy mate puzzles and tackled every day a few hundred. At the same time i calculated my progress on this set. I did not get any quicker. My performance did not increase. I did check the relationship between my solving speed and seeing the same puzzle a second time again. Interesting was, that my speed was unchanged if i did not see this puzzle for 8 days.  So i made sure that i see every puzzle less than 8 days again and my performance did increase. Seemingly there was a transfer to other problems too, but the statistic was very small and the salt was dry, dry , dry

Munich made salt in a slightly different way and he is one of the few real improvers in tactic i know , but he had to make a lot of salt and his improvement was not really gigantic.

So i did hope to bypass the salt mines somehow, get to the same result more intelligent. I still think that the right "motor skill" can work too. If we could invent to "every" sub skill enough exercises... But even that is quite salty.

Analyzing the difficulty of improvement in tactics i was looking for the difference in "improvable" and "not improvable tasks". A task is improvable if you can reach a level of performance of a Chess-Master. A lot of tasks are seemingly improvable. Temposchlucker did report of a knight move puzzle where he was able to perform at master level because he did make many exercises in the game of Troyis before. I did exercise the board vision exercise "attacker" until i reached a level of a master.

So if we can find enough improvable sub skills we might break the improvement barrier ?

The easiest not improvable task i found is "Mate in 1". Tomasz did experience, that solving 50 000 Mate in 1 puzzles don't improve his solving speed "decisive". I already had the experience that simple mate puzzles are "not improvable" so i did expect his experience.
It is of course astonishing that such a tiny little subset of tactical puzzles is already so difficult to handle. If you cant improve at such a small subset of tactics in 50 000 attempts, then you can you expect to improve at a tactic server with so many different pattern in 50 000 attempts?

Now Tomasz try to break his Mate in 1 - plateau by improving sub skills like : Find all checks. I'm sure he will find different exercises and i hope he find a method this way.

But i will try again to improve by repetition. I did create sets with 300 mate in 1 puzzles and i try to improve my speed at these sets...and hope i might get quicker at other, unseen mate in 1 puzzles by doing so.
At the moment i work on 6 of these sets, that are 1800 puzzles. My speed did increase by 10% at these sets. Happy me, i have more than 200 000 puzzles left to analyse any transfer. But first i have to develop ideas what and how to continue.

But i can tell you: salt is dry, dry, dry

I still don't understand, why such a method might work. The only explanation i could think off  is, that the access to the memory might get improved. There are somehow related exercises in the IQ improving scene. Pattern search exercises to improve thinking speed. W'll see.