Monday, September 18, 2006

This is working. Minicircles 1-10 Done

I have worked through Phase 1 of Chess Tactics for Beginners 10 times. I am finishing the 120 mate-in-one problems in 10-15 minutes. Typically I see the solution instantly, but sometimes it takes up to five seconds for the answer to click into place.

More exciting, in blitz games mates are starting to pop out at me in ways they didn't before. I have won a bunch of games in the past week because my opponent missed my mate-in-one threats. Often they were weird mates, perhaps ones I would have missed before this training. I have also started to see my opponents sneaking up on me with mating attacks that I wouldn't have seen even a month ago.

There is further evidence that it is working: after my sixth minicircle, when it was clear I had learned the problems cold, I went to a new set of mate-in-one puzzles to see if I had merely learned the 120 problems, or if learning them actually improved my ability to see other mates. On average, I saw the mates pretty fast, usually within 10 seconds, suggesting that my mate vision is not a brittle artifact. (Note that in the statistical pattern recognition literature, you always divide your data into a 'training set' and a 'test set': after using the training set to set the parameters on your mathematical model, you then apply this model to your test set to see if it generalizes beyond the training data, as there is a danger that your model has simply 'memorized' the training data. We usually call this a 'brittle' model because it is easy to 'break' it).

Many of you are thinking "Wow, so you've reached the level of crappy patzer." I can only respond with an embarassed shrug. We'll see if it affects my performance in real (i.e., long) games.


# CirclesPercent Correct
Problem Set 11098-99-100-100-100-100-100-100-100-100
Problem Set 20
Problem Set 30
Problem Set 40
Problem Set 50
NOTE: Circles done with CTB.

9 Comments:

Blogger Jim said...

Well done! This was exactly my experience!

Can we say the conclusions I reached about massively repetitive training is no longer anecdotal, but based on empirical data?

I'm looking for recognition here, mind you. I gotta find a way to make money of these blogs somehow [grin]. . .

9/18/2006 05:25:00 PM  
Blogger Blue Devil Knight said...

J'adoube: I think you should trademark the term 'Circles of Death' before de la Maza publishes his second edition!

9/18/2006 06:38:00 PM  
Blogger Loomis said...

How do you decide how many mini-circles to do? Do you have to do them at 100% a certain number of times before you move on?

9/19/2006 01:41:00 PM  
Blogger Blue Devil Knight said...

Loomis: I am flying by the seat of my pants. Basically, when it is clear I know the problems really well, I move on to the next problem set.

So, I am not doing them every day anymore. I'm in the middle of the second problem set, which will probably take 2-3 weeks the first time.

I'll review the first problem set when it seems prudent (every few days). Since I now finish the first problem set so quickly, it isn't all that big a hassle to repeat it.

It will be very interesting to see how performance drops as a function of days since rehearsal.

9/19/2006 04:29:00 PM  
Blogger Sancho Pawnza said...

Glad you guys are finally coming around to tackling smaller blocks of problems, and using mate problems to boot! I’ve said it before; mating patterns/threats are the best mainly because they have to be answered. Not only can you win the opponent’s King but you can usually pick up lots of extra material just with the threats alone!

Smaller problem sets are the best way to build pattern recognition in my opinion.
But don't make the same mistake I did when it comes to ignoring the value of calculation muscle.
Pattern recognition and calculation muscle are two separate things entirely, and have to be trained using different methods.
I covered some findings in my March 9th 2005 post.
But basically the pattern recognition can be learned very quickly if you focus on small blocks with a periodic review after completion just to keep it fresh.
When training the calculation muscle you should take your time with each problem and think the solution through to completion. The number of problems isn't as important as getting the correct answer regardless if you can only solve a few each day. I would be more inclined to pick theme based problems if you are just starting out (i.e. deflection, removing the guard, sacrifices, 2-fold attacks, etc.) this is where CT-Art actually works well because it can be set for theme training and the problems contain enough flaws to make you find your own answers with your own analysis.
You could then return to the same problems at a later date and convert them into "pattern problems".

Playing through master games covering up the moves is also a good way to work on your calculation, use a book with good annotations so that you can compare notes with the author.

I'm glad to see the mate patterns are starting to pay off.
Wait until you hit the mate in 2's-3's and get exposure constructing mating nets. That's when the real fun begins!

9/19/2006 09:22:00 PM  
Blogger transformation said...

blue, congrats on hard work, and your fabulous blog.

short and sweat this corner: im back on CT-ART daily, as is takchess and samurai (oddly, we all seem to be at level three+ or four) and 450 to 500 problems, if i recall correctly. we all are in the torture chamber nearby. you are not alone. even if chains have a different tincture.

20,000 CTS tonight. 84.7%.

hope the south is not too hot now!
dk

9/20/2006 02:44:00 AM  
Blogger Blue Devil Knight said...

Sancho: it is good to emphasize that calculation and pattern recognition are quite different species of fish. To work a bit on my calculation muscle, in my first few passes through the minicircle problems I really take my time to think them through. Then, I try to preserve that accuracy while moving faster each time, just as recommended by MDLM. I think this is a good feature of the MDLM program, which I'm preserving. The main difference is that I have picked a smaller problem set and repeat it lots (based on you, J'adoube) before moving on, and don't stop at seven, but am learning them like I know 2+2.

9/20/2006 09:27:00 AM  
Blogger Temposchlucker said...

Tonight I had a closer look at the Grand Prix Attack with 1.e4 c5 2.Nc3 and 3.f4. The reason for this is that I never had good results with the Morra Gambit (1.e4 c5 2.d4 cxd4 3.c3, I always had the feeling I was attacking at the wrong side of the board i.e. the queenside) nor with the Alapin (1.e4 c5 2.c3 which is very slow and offers black too much possibilities for trading off pieces).

It looks very playable. The positions have some resemblance with the Kings Gambit declined and with a reversed Dutch defence, which are positions I'm familiar with. So I'll give it a shot this season. Thanks for the tip!

BTW I used the book of Hodgson and Day.

9/20/2006 06:30:00 PM  
Blogger Blue Devil Knight said...

Cool, Tempo. I'll be curious to see how it works out. I really like the GPA too. It just feels comfortable for me, partly because I often transpose into the King's Gambit Declined from my Bishop's opening...

9/20/2006 08:30:00 PM  

Post a Comment

<< Home