TalkChess.com

For the rest the Crafty book also gets stuck for White after 1.d4 Nf6, and then is going to think for itself it seems. and comes up with 2.Nf3 or c4 after a long time. they should fix that i think.

Norm Pollock wrote:
I did realize that. But what I am asking for is a way to capture the games that were excluded. Namely, the games that seem to have an incorrect result based on the evaluation of the final position.

-Norm

Don wrote:

Richard Allbert wrote:I agree with Harvey, the evaluation of each position is going to make the whole thing last a long time.

I wrote a bookmaker for Jabba - what I did was hunt down all the Playchess Engine room pgn files I could find - as in the engine room there you basically have the top engines playing vs themselves with different books.

I found out here what Bob wrote - the IO is the bottleneck.

I then computed win / loss / no. games, and simply wrote the positions in the book according to thresholds for win and number of games.

Problem is Jabba has larger weaknesses than the book, so I'm not sure if the result was a good book or not

Another method could be to take 150 ECO positions (there was an epd file posted here with them) to about move 6 or so, and play very fast matches with good engines from these positions, and build a book from the result.

Initially, I made the mistake of using a massive human games DB, unchecked, assuming that the bad moes would be filtered out by the "add to book if > X games played) and watched in horror as Jabba blundered a piece in the opening - a blunder that had apparently occurred enough times to make it into the book!

Good luck

Richard

In the book I talked about I was horrified to see that 1. h4 made it into the book. It was played at least 10 times. The TWIC database has mostly games by strong players but it also includes important tournaments that can have very weak players in them such as tournaments for the very young (where the strongest players may not even be masters and and the weakest may be quite weak or even beginners.) I did not filter by ELO.

I manually removed 1. h4 but didn't try beyond that - most the blunders are probably in the first 3 or 4 ply but I'm sure there are a few beyond that which still survive the "at least 10 time" frequency test.

jwes wrote:

Don wrote:

Richard Allbert wrote:I agree with Harvey, the evaluation of each position is going to make the whole thing last a long time.

I wrote a bookmaker for Jabba - what I did was hunt down all the Playchess Engine room pgn files I could find - as in the engine room there you basically have the top engines playing vs themselves with different books.

I found out here what Bob wrote - the IO is the bottleneck.

I then computed win / loss / no. games, and simply wrote the positions in the book according to thresholds for win and number of games.

Problem is Jabba has larger weaknesses than the book, so I'm not sure if the result was a good book or not

Another method could be to take 150 ECO positions (there was an epd file posted here with them) to about move 6 or so, and play very fast matches with good engines from these positions, and build a book from the result.

Initially, I made the mistake of using a massive human games DB, unchecked, assuming that the bad moes would be filtered out by the "add to book if > X games played) and watched in horror as Jabba blundered a piece in the opening - a blunder that had apparently occurred enough times to make it into the book!

Good luck

Richard

In the book I talked about I was horrified to see that 1. h4 made it into the book. It was played at least 10 times. The TWIC database has mostly games by strong players but it also includes important tournaments that can have very weak players in them such as tournaments for the very young (where the strongest players may not even be masters and and the weakest may be quite weak or even beginners.) I did not filter by ELO.

I manually removed 1. h4 but didn't try beyond that - most the blunders are probably in the first 3 or 4 ply but I'm sure there are a few beyond that which still survive the "at least 10 time" frequency test.

A simple fix would be to add a second condition that a move is played in a given position at least n percent of the time. This could eliminate moves near the starting position that are rarely played.

Norm Pollock wrote:

jdart wrote:To save the output you need to redirect to a file, for example:

playchess2 input.pgn > output.pgn

--Jon

I did realize that. But what I am asking for is a way to capture the games that were excluded. Namely, the games that seem to have an incorrect result based on the evaluation of the final position.

-Norm

Dan Honeycutt wrote:

Norm Pollock wrote:

jdart wrote:To save the output you need to redirect to a file, for example:

playchess2 input.pgn > output.pgn

--Jon

I did realize that. But what I am asking for is a way to capture the games that were excluded. Namely, the games that seem to have an incorrect result based on the evaluation of the final position.

-Norm

Hi Norm

My book utility is at a rough beta stage at the moment. It has a pgn check function which looks for mismatched braces, conflicting results and the like. It also has my Cupcake engine built in. I don't think it would be too much trouble to make a "suspicious results" check if you want. Do you have criteria you'd like to use?

Best
Dan H.

Norm Pollock wrote:Hi Dan,

Thanks for the response. For a start, you could try:

1) If the game ended in a draw, in most cases, the pawn-based score for the final position should be between -1.00 and +1.00 inclusive regardless of the color to move.

2) If the game ended 1-0, in most cases, the pawn-based score for the final position should be greater than +2.00 if White is to move, or less than -2.00 if Black is to move.

3) If the game ended 0-1, in most cases, the pawn-based score for the final position should be less than -2.00 if White is to move, or greater than +2.00 if Black is to move.

If the result is not consistent with the score of the final position using the above criteria, then the game should be extracted for analysis. There could be countless reasons for the mismatch- such as time pressure, blunder, a non-winnable position with unequal material, or a recording error.

-Norm