New idea for a chess puzzle book, with two samples

[dkappe]

I’ve always enjoyed the “what’s the next move” sorts of books. I do have some issues with them, though. First, they get you to focus on only one move, the “best” move. Second, there are not enough of them.

I came up with a three move quiz a few years ago, where you pick/rank your three best moves and get 3 points if the move played matches your top move, 1 point if it’s one of your other two and zero points if it doesn’t match any.

With the advent of neural net engines, the quality of computer chess advice has improved. I’ve now taken to generating quiz books for three move, first by using sf11 to filter out ridiculous positions (mostly), then picking a few positions and analyzing them with Fat Fritz and SF11. The “best” move is the one recommended by FF.

I have two samples:
- positions from the black side of the Najdorf by Loek van Wely
- positions from the white side of the Scotch by Sergei Rublevsky

https://www.dropbox.com/s/3ynylumdqt5b8 ... f.pdf?dl=0
https://www.dropbox.com/s/s56yx2gke0c0e ... h.pdf?dl=0

Would appreciate any feedback.

[Ovyron]

The feedback I'd give is about Murphy's Law: How can you avoid positions where both Fat Fritz and Stockfish fail to give the best move? For those the best move would not give the best score.

Best move quizes work because eventually all entities agree that there's a best move (and sometimes those are "cooked" because it turns out it was wrong, or there were many), if there's multiple moves that can be played, it's unclear why one should give more points than another.

Someone that has a move that is better than what SF11, Fat Fritz and the human played yet gets 0 points would feel bad.

[dkappe]

The feedback I'd give is about Murphy's Law: How can you avoid positions where both Fat Fritz and Stockfish fail to give the best move? For those the best move would not give the best score.

That’s an issue also with the original “best move” books: Alekhine — or whoever — didn’t always play the best or only best move. For now we’ll just have to feel bad from time to time.

[Ovyron]

Ah, I see, yours is an improvement over a system that already had the flaw.

[Deberger]

This looks like a laborious and transparent advertisement for Fat Fritz.

[dkappe]

This looks like a laborious and transparent advertisement for Fat Fritz.

Looks can be deceiving. I assume you are implying I have a financial interest in Fat Fritz? I think there’s a corollary for computer chess to Kissinger’s famous quote about academe: “Why are the fights in academe so dirty? Because the stakes are so low.”

I do this for myself and thought I’d share. I have some experience with leela chess and various nets and have selected Fat Fritz because it works best for me.

[Ovyron]

It would be very interesting if you could compare positions where Fat Fritz and Leela produced something different and Fat Fritz was better, as those would work for justifying using commercial over free software.

[dkappe]

It would be very interesting if you could compare positions where Fat Fritz and Leela produced something different and Fat Fritz was better, as those would work for justifying using commercial over free software.

Since I’m making the books available for free, there’s nothing to justify. Use them or don’t.

[Deberger]

It would be very interesting if you could compare positions where Fat Fritz and Leela produced something different and Fat Fritz was better, as those would work for justifying using commercial over free software.

One could simply assert that Fat Fritz is better than Leela and Stockfish, while creating an elaborate framework to reinforce that belief.

[Ferdy]

I’ve always enjoyed the “what’s the next move” sorts of books. I do have some issues with them, though. First, they get you to focus on only one move, the “best” move. Second, there are not enough of them.

I came up with a three move quiz a few years ago, where you pick/rank your three best moves and get 3 points if the move played matches your top move, 1 point if it’s one of your other two and zero points if it doesn’t match any.

With the advent of neural net engines, the quality of computer chess advice has improved. I’ve now taken to generating quiz books for three move, first by using sf11 to filter out ridiculous positions (mostly), then picking a few positions and analyzing them with Fat Fritz and SF11. The “best” move is the one recommended by FF.

I have two samples:
- positions from the black side of the Najdorf by Loek van Wely
- positions from the white side of the Scotch by Sergei Rublevsky

https://www.dropbox.com/s/3ynylumdqt5b8 ... f.pdf?dl=0
https://www.dropbox.com/s/s56yx2gke0c0e ... h.pdf?dl=0

Would appreciate any feedback.

How about showing the top 3 moves of FF or SF too? Then give points based on the score of those top 3 moves.
If top 1 move has a score of +0.15 (15cp) and top 2 move has a score of +0.10, since the scores are close, top 1 and 2 moves will get 3 points.
If top 1 move has a score of +0.15 and top 2 move has a score of -2.0, top 1 move will get 3 points, top 2 move may get 0 points because it is losing.

Perhaps the maximum points for top 1 move can be made 100. If according to the engine the top 1 move has a score of +1.5, and could translate to a score percentage of 80% according to some data for this engine, then the points for that move can be made 80 points as well. So move point system can be easily calculated.

One idea is to classify the top moves. Does the move seeks complications, does the move seeks simplification or something of a style, like Tal style, Capablanca style, Carlsen style etc. Once the tests are completed, a player can be profiled i.e 70% likes to complicate, 30% likes to simplify, or 60% Tal, 30% Capablanca, 10% Carlsen.