Thursday, August 31, 2006

Do You Count Shades of Red?

Silly Question. You can count sheep but not wool. You can count socks but not colors of socks. You can count cars in traffic but not the danger of an accident.

My Target Feature Count is the silly attempt to execute an excellent plan. Thanks, Tempo, for your comment and posts, they helped me a lot to come to this conclusion.

Nobody can deny that finding and evaluating targets in a chess position is an excellent plan. In order to distinguish targets from non-targets we need features to look for. Up to this point everything is okay. But the idea of counting these features must be silly, because target features are sort of colors or flavours, and these cannot be counted. Can you imagine a wine taster counting the flavours he perceives?

I have a better idea to deal with targets: upgrading and downgrading them as we look longer and deeper into a position. Just imagine every piece and pawn has small red and green lamps on its top. If a piece is safe, the green lamp is on. If it becomes less safe, the green light dims down and the red lamp dims up. If a piece becomes a target, the red light brightens. If the king is one move from checkmate the red light flashes and an alarm bell rings.

In principle, chess tactics is very simple. All you have to do is to look at the targets, sort them by brightness of their red lamps, exploit the one or two top targets and forget the rest. That's it.

CTS performance:
8-30: 91% @ 1460 (n=100) :-)
8-31: 88% @ 1465 (n=190) :-(

Wednesday, August 30, 2006

Pattern Search Tree

It is interesting that both Tempo and I are dealing with the same question these days: How many different tactic patterns exist? Our estimates differ a lot, from my initial idea of over a thousand pin patterns to his idea of pin as a single pattern. This could be viewed as a source of dispute, but it is not. I even go beyond Tempo and postulate that pins are only a fraction of one and the same super-pattern.

Before I explain this let me tell you how I found out. I was jogging this morning with my old doggie, I have to jog very, very slowly so that she can follow me (years before it used to be the other way round). I just was contemplating what it could mean to count patterns, and if there is any method that would allow a reasonable count. And then it flashed to my brain. Pattern search is a kind of tree search, and the number of patterns you get depends on the level of ramification.

Thus, Tempo is right to count one single pattern of the pin, and I am right to count 144 different pin patterns. We just count on different levels.
  • The first level of ramification is the dichotomy of «tactical» vs. «non-tactical» positions.
  • The second level is the super-pattern or pattern category, such as «inline-patterns» (pins, skewers, forks of B, R and Q on a single file, rank or diagonal)
  • The third level is the pattern itself, for instance the «pin»
  • The fourth level is the most important feature of the pin. If a queen is pinned against the king, this feature will be the «pinnee value», if the pin helps a mate, it will be the «pin degree» (absolute pin required)
  • At this level, additional features or details will not be helpful. Instead, the patterns of achieving will emerge: How do I get this pin ready? What pieces should be pushed or pulled to what squares by what means?

Instead of doing more pattern counting my next posts will deal with building a pattern search tree. The idea behind this project is the way we detect simple, well-known patterns in noisy and confusing environments of positions we see the very first time. The first thing we see is two targets, such as the King and an unprotected piece on the same line. The «inline-alert» lamp flashes up. Then we find an attacking piece that can go in front of the target of lesser value (pin), of the target of higher value (skewer) or between them (fork).

My idea is that we can detect patterns faster if we have a clear concept of what exactly we search and how these things are correlated. My future performance at CTS will show me if I am right.

(Just for the record: Yesterday I did about 200 CTS problems with about 90% success rate at an average rating of 1450.)

Tuesday, August 29, 2006

A Second Look At Pins

I am not very happy about my post of yesterday. It has a fundamental weakness and misses at least one important point.

For instance, it cannot be very helpful to mix patterns and the tactical operations to achieve them. There are not 1152 pin patterns, but only 144 (3x2x2x3x4) with a number of forcing methods, probably more than eight.

Also, my post missed the point that a pin can be a pattern in itself, for instance, winning material by a pin, or on the other hand it can be a method of removal of the guard, for instance, allowing a mate. I think it is very important to distinguish these two kinds of pins, because the weight of the features are different: A removal of the guard pin should be absolute (in a mate it must be absolute), but the value of the pinnee does not matter. Therefore:
  • 144 basic pin patterns (3x2x2x3x4, see post of yesterday)
  • 12 pinning the guard patterns (3x2x2)

Monday, August 28, 2006

1152 Basic Pin Patterns

After my first expedition into the pattern jungle and the encouragements of my fellow Knights I am kind of frightened about my own courage. I fear to get lost in the jungle, ending up with a lot of work without any useful results.

Before doing more such expeditions I need a kind of map. The first pattern to map are the pins, for various reasons. Weeks and months ago I missed pins at CTS over and over again. Then I becan to focus on pins. I got the pins on files quickly, but still missed diagonal pins. I focused on diagonal pins and missed pins on ranks. Therefore, the first and most important feature of a pin must be its direction. Here is my classification of pins (the actors are Pinner-Pinnee-Target):
  1. Direction: Files, Ranks, Diagonals (3 possibilities)
  2. Degree: absolute (Target=King), relative (2 possibilities)
  3. Pinner-Pinnee-Relation: Pinnee may capture or may not capture pinner (2 possibilities)
  4. Pinnee Value: Pinnee is of higher, of equal or lesser value than pinner (3 possibilities)
  5. Pinnee Exposition: unprotected, not sufficiently protected and may be once more attacked, protected but may be further attacked by a piece of lesser value, sufficiently protected (4 possibilities)
  6. Creation of the pin: unforced (pre-existing), pulling the target into the pin by a sacrifice or by a capture-recapture sequence, pushing the target into the pin by a threat or by a check, pulling the pinnee by a sacrifice or by a capture-recapture sequence, pushing the pinnee by a threat (8 possibilities)
So there must be at least 3x2x2x3x4x8 = 1152 basic pin patterns.

Wow, I definitely must have underestimated the number of basic patterns when pins alone make such a huge number. And no wonder I used to miss so many pins earlier. Of course the number of basic pins is too large to store them all in the memory. But the pin features or, as they could be called, dimensions should be stored there for a quick retrieval.

Sunday, August 27, 2006

How Many CTS Patterns?

Today I started a project that should allow me to estimate the number of different patterns at CTS. My first session, of course, does not give me any order of magnitude yet. But, as I hope, with more sessions this should be possible. I had decided to have a closer look at each problem after having done it, but I could not help looking closer while solving, so I lost about 30 rating points. But this does not bother me at all, because I'll gain them back later.

I did 50 CTS problems, 49 right, 1 wrong, on an average level of 1415. I identified and listed every pattern. The 50 problems presented a total of 69 patterns or 42 unique patterns. This means that every pattern occurred roughly 1.5 times on average.

Here is the preliminary pattern hitlist, ranked by number of occurrence. The first one was presented 5 times, the last one twice. I hope the names to be more or less self-explanatory. The mates are always from the eight basic mates presented by King of the Spill.

Capturing Guard
Capture With Threat
Chasing Guard
Discovered Attack
Chasing Guard With Check
Frontal Queen Mate
Pinning Piece of Higher Value
Bishop Fork
Oblique Queen Mate
Overworked Knight
Promoting to Queen
Queen Fork
Rescue By Trade
Skewer
Trapping Bishop

Saturday, August 26, 2006

Guard Swapping And More Basic Patterns

There is a pattern at Chess Tactics Server that I get right about 100% in an average time of about four seconds. Whenever it appears on the screen, I hear my inner voice say, hello again, here we are, and ... gotcha! yeah!

I have always asked myself why I get this one faster than most other patterns at CTS. I am not sure if I am right, but I have an idea: It is because I can describe this pattern in words that make flash up multiple pictures in my inner eye.

And this is the guard swapping pattern: The problem attacks my unguarded queen with its guarded queen. I capture the problem's queen guard with my own piece that becomes a guard of my own queen. Now mine is guarded and the problem's is no longer guarded. I have won a piece.

Of course there are hundreds of possible variations of guard swapping. The target piece may be a queen, a rook or any other piece. But mostly it is a queen, because CTS loves queens. The problem's guard may be any piece, and my attacking and guarding piece may be a knight, a bishop, a rook or even a queen.

I recognize guard swapping quite fast, whatever the position looks like. I think this is so because I have stored the basic pattern in my mind, and it is stored in a way that allows me to recognize it fast in multiple variations.

Could this be a concept of making progress at CTS? I do not know. But I suspect that it might be. There are tens of thousands of problems at CTS. There are thousands of pattern variations. But I believe that there is only a limited number of basic patterns.

Just take mates. All possible mating patterns can be reduced to just thirteen basic mating patterns, as King of the Spill has shown in his brilliant post more than a year ago. Maybe I am wrong, but it is possible that it is sufficient to have stored all these basic patterns plus the ability to recognize blocked squares of any kind, be they occupied by pieces of the defender or under control of attacking pieces or beyond the edge of the board (for example in back rank mates).

Just take the overworked piece. If you have stored this basic pattern, then you see it on the spot. You do not have to learn all possible variations and permutations of the guarding piece (K, Q, R, B, N, pawn), first target (K, Q, R, B, N, pawn), second target (K, Q, R, B, N, pawn), first attacker (K, Q, R, B, N, pawn), and second attacker (K, Q, R, B, N, pawn). I am no mathematician, but I guess that we speak of roughly 6x6x6x6x6 = 7776 possible variations of the overworked piece, but all these are just one basic pattern.

The example of guard swapping may show me a way to improve: First recognize a basic pattern, then describe it in words, then give it a name that serves as a memory handle, then imagine this pattern on occasions such as waiting for the bus, before falling to sleep, while shaving in the morning, and so on. I have no idea of how many basic patterns exist. But I am quite sure that a hundred of the most important ones will cover the majority of CTS problems or tactical occasions in OTB games.

Friday, August 25, 2006

Quick Eye, Slow Hand

Tempo's comment on my last post tells me that I have to make certain things clear. That is, the difference between a quick eye and a quick hand. Both may be labelled as speed, but this is very misleading.

I do not remember who gave this advice: When you play chess, you must sit on your hands! If I would have to choose the ten golden rules of chess, this one would be among them, this is for sure.

I completely agree with Tempo that a quick eye is the key to tactical success. It is also clear for me that a quick eye is the key to accuracy. Just because the more patterns you see per amount of time, the smaller the danger that you miss an important one.

If there is any difference in our approaches to CTS, then it is the time required to actually move, or the quick versus the slow hand. In my last post I have compiled seven reasons for choosing the slow hand approach. Today I think that matters are much simpler. There is only one reason:

Not training the bad habit of hope chess at CTS.

Thursday, August 24, 2006

Different Approaches at CTS

It is interesting to see how different the approaches of various Knights at CTS are. I am quite happy about that, because it forces me to think about what I do, and why I do it, and if I really should do it.

Tempo, a leading member of the heavy-repetition-at-high-speed school at CTS, has just posted about his reasons and the conceptual backgrounds of his strategy. Surely his method works for him, for his rating is more than a hundred higher than mine, and it was not when he began at CTS. I saw him coming from behind, catching up and leaving me back.

Of course, sixty thousand repetitions are not for nothing. And he says that this is not all, he does a huge number of non-rated repetitions besides. Wow.

I belong to a different school. We may call it accuracy-over-speed or success-oriented. I'll try to give some comments on my reasons for this choice. BTW, it is relatively new, because up to 20k repetitions I tried to optimize my rating, slowing down only after my success rate dropped too fast.

Firstly, my goal is not training for blitz, but for slow OTB games. This favours accuracy, because in a slow game it is very rare to have only 3 to 10 seconds per move left.

Secondly, the patterns at CTS are quite simple, sometimes even very simple, as Tempo says. That is, they can be found very quickly, even at very first presentation of a problem. But before seeing them quickly, you must see them at all. Therefore: first accuracy, then speed up.

Thirdly, every wrong move is a wrong engram into the memory. Therefore I try to avoid it as much as possible.

Fourthly, the reason of my OTB rating plateau, despite of heavy tactical training, is lack of accuracy, or, more precisely, lack of security. In principle, I am quite good in tactics if I KNOW that there is a tactic, but in games I constantly miss tactics, because I miss simple tactic features (or targets) such as underprotected, exposed or trapped pieces.

Fifthly, I am convinced that you must sow before you can harvest. Thus, if I try to optimize my rating during training, the effect will be less than optimal. I think I do not differ from Tempo in this question, because his training is heavy unrated repetitions. I did this before, but now I find it more interesting to have the constant challenge of being rated. Just like in a game.

Sixthly, an important goal of my training is to distinguish patterns from noise, and so I am quite happy if I am presented problems I never have seen before. By noise I mean such things as winning a queen if there is a mate in one. Even a mate in two is noise against a mate in one (and is punished by CTS). You cannot do such a training if you go for speed.

Seventhly, and there may be many more reasons, just this one: Looking at every target on the board, not missing a single one, looking at targets in the right order of importance, spotting leading features of a position such as quickly seeing if kings or queens are in danger or in safety, and the like. All this is possible with an a-tempo approach only if you have done it more slowly before, time and again.

Today, I had the best CTS session ever: 93/97 correct (96%), with a winning streak of 53 correct in a row, at an average rating of 1442. Only one of the fourteen tacticians with 95+ percent has a rating of 1500+. And yeah, Tempo, we are neighbours in the success ranking now.

Wednesday, August 23, 2006

Garry And Me

Before a game, he carefully adjusts all pieces exactly on the centers of their squares. So do I. He turns the knights so that they look to the opponent. So do I. I wish we had more in common ...

The event was great yesterday. A huge crowd. Very spectacular games. I guess it was the last time ever to see these four legends together.

Let's begin with Victor Korchnoi. The time schedule has been changed in the last minute, he had agreed to play a rapid tournament and then was told it was blitz (5 minutes plus 3 seconds per move). He made the best of it, knowing he had nothing to lose, he refused all draw offers and delivered an epic fight to his old rival Karpov in a game with no less than five queens, four of them simultaneously on the board. I have great respect for this old fighter who ended with just a half point.

Judit Polgar who was there with her baby showed spectacular aggressive chess, against Kasparov she sacrificed a knight in the opening. Great tactician, this lady. An inspiration for us Knights.

Karpov played solidly as ever. Played very slowly in the openings, not fearing time trouble, and in the end he showed his defensive brillancy, finding the right moves in seconds. Great respect for him, too.

But Garry is still a class of his own. The commenters were excited about the harmony of his pieces, working together in an optimal way. And he was the only one who won all his 20 simul games. Garry, come back to chess and forget politics!

Here is a report in German with a lot of pics. And here are the games.

Oh, I nearly forgot to mention that I had no luck in getting a simul place.

Tuesday, August 22, 2006

Chess Champions Day

Today I'll be in Zurich: Garry Kasparov, Anatoli Karpov, Judit Polgar and Victor Korchnoi will play a rapid tournament and a simul at 4x20 boards, sponsored by Credit Suisse. Four legends in one place, just one hour from my home, so I must be there!

I had participated in a contest to win one of the simul seats. There were one tactical problem from a Polgar game and three questions of chess history. The tactics was no problem, but I was too bad in history, and I had no luck in the drawing from the false answers. Nevertheless, I'll try to see if one of the winners doesn't show up so that I might take his seat.

Anyway, the rapid contest will be very interesting. All games will be transmitted to a room where a grandmaster will give his live analysis. Of course I expect Garry to win, and I hope that old Vic will not take the last place. I admire the Polgar sisters, so I hope Judit will end as second.

All games will be broadcast in the internet.

Monday, August 21, 2006

80 Percent at 1452

Yeah. Passed the 80% milestone at CTS yesterday, in a session with roughly 90% at an average rating of 1450. Since I have made my move only if I am fairly sure that it is correct, my success rate has improved by more than two percent now. Also I have gained back 20 rating points.

Update: Today I did a hundred problem session with 94% success at an average rating of 1440.

Saturday, August 19, 2006

Silent Errors

Bad luck! You failed to solve ... This message on CTS tells you that you have an opportunity to learn. The trick is to find out what you did wrong and what you should do better next time. So far so good. But errors are much more frequent. They even occur when you win against the problem. They may even occur if you win fast and you gain rating points!

You do not believe me? Just a common example. You see a good move. You wait for the move of the problem. You are about to do what you have seen before. Then, oops, you see that your king is in check. You parry the check and then make your winning move. The silent error is that you did not check king safety first.

I have found that I lose rating points mainly because of such silent errors:
  • not checking king safety first
  • not checking queen safety second
  • looking too long at patterns that do not work
  • stopping to look for more unprotected pieces after I have found one
  • ...
Conclusion: CTS is an excellent tool for training of thought process on the micro-level, that is, in the millisecond range. It is important always to do the most important things first, no matter how beautiful the patterns that jump into your eyes. You can win a queen? Forget about it as long as you did not look for a mate! In most cases the six seconds before the problem makes its move are enough to decide if you can go for mate or not.

Monday, August 14, 2006

Defense and Attack are the Same

Thinking defense is thinking attack for the opponent. Very important, and one of my main weaknesses. But I have started to work on this. CTS is an excellent tool for defensive training. The problem is that defense needs extra time. My rating has dropped to around 1430, but I do not care for the moment. My success rate is around 90% in the sessions now, and I have gained one percent since I changed my strategy, up to 79.1. Way to go!

Sunday, August 06, 2006

Started Safety Training at CTS

After my detection that safety is my main weakness despite of some tactical skills, and that CTS is excellent for safety training, I have switched again to a strategy aimed at a percentage as high as possible, no matter how much rating I may lose.

Today I did 100 problems, solved 96 and failed 4, and I lost 13 rating points. My longest winning streak was 37 in a row, losing 6 rating points.

My overall success rate increased from 78.2 to 78.3 in this session.

Saturday, August 05, 2006

CTS and TFC

CTS is Chess Tactics Server. TFC is Target Feature Count. The two have a lot in common. In CTS you must find targets as quickly as possible. That is, exposed kings, unprotected or underprotected pieces that can be attacked, trapped pieces for material win or trapped kings for mate, hanging pieces to be taken or rescued. In other words: You must spot and evaluate target features first and only then proceed to candidate moves.

Thus, I have found a second meaning of CTS: Candidate Target Search, and, after this is done, Candidate Target Selection.

Also, both CTS and TFC are about safety. The good thing about CTS is that you never know who is unsafe, you or the problem or both. You constantly have to be alert. I think CTS training is the best you can do for blunder prevention.

At the moment, my chess training is divided between CTS and postmortem analysis of games (own and master games) using TFC. At CTS I try first to improve my success rate and then my rating.

Friday, August 04, 2006

Making TFC Fit For Games

A game is no scientific study, and the analysis as I did it in my test is too artificial for practical use. Simple cases such as a piece hanging or en prise, a king in check or threatening mate in one do not require a TFC, because it is clear that these are significant targets. TFC is useful for detecting non-obvious targets. Most significant targets have a TFC score of three and more. But it does not matter if it is 3, 4 or 5, because the position is much more important.

Therefore, I classify targets as TFC positive (score 3 or more) and TFC negative (less than 3). With every move, on average, one or two new targets emerge, and a similar number disappears again. Their lifespan is very different, ranging from one ply to twenty or more moves. On average, four or five targets are present at any moment, but in some highly tactical positions there may be more than ten.

Conclusion: TFC does not eat away much time, and it does only what has to be done anyway in the safety check of every move.

Wednesday, August 02, 2006

Scientific Test, Part VII: Conclusion

Target Feature Count (TFC) may be a useful element of a move search heuristic. This has been a preliminary test. Further examinations are needed to answer the following questions:
  1. What is the best method to separate significant from non-significant targets?
  2. Is the TFC score useful for ranking targets or only for candidate target selection?
  3. How can positional features be added to the TFC list?
  4. How can target features be weighted?
  5. What is the significance of TFC in the opening, in the middlegame and in the endgame?
  6. Are the number of targets and their TFC scores different in open, semi-open and closed games?
  7. Is TFC useful in quiet, positional situations?
  8. What is the typical life of a target? When does it emerge? How long does it live? When does it disappear?
  9. How many significant targets are present in typical situations of a game, and how do they score in TFC?
  10. How can TFC be optimized for use in practical games? What amount of time is required? How can TFC be done under time pressure?