Silent but deadly

[Jouni]

Interesting suite Dann! I run it with Rybka and it solved all except 3 with 10s level (P2400 only). But when suite is debugged with Rybka no surprise
Interestingly Toga and bright solve most positions with 1 ply, but Rybka or Fritz have always 5 resp 8 plys - probably they have some internal minimum depth for display.

Jouni

[Zach Wegner]

Thanks very much Dann. I guess you made this test at my suggestion, but I had no idea you would make such a nice testset so quickly. The reason I didn't respond earlier is that I broke EPD support with one of latest changes. I'll post my results later.

[Tord Romstad]

Thanks very much Dann. I guess you made this test at my suggestion, but I had no idea you would make such a nice testset so quickly. The reason I didn't respond earlier is that I broke EPD support with one of latest changes. I'll post my results later.

You do have UCI support, don't you? Then you have EPD support, even if you don't think you do.

Try this something like this:

Code: Select all

polyglot epd-test -epd sbd.epd -min-time 10 -max-time 60

There are also switches named -min-depth, -max-depth, and -depth-delta. It should be easy to guess what they do, and it's also not hard to read the source code.

Tord

[Zach Wegner]

You do have UCI support, don't you? Then you have EPD support, even if you don't think you do.

Haha, my EPD support actually got broken when I was fixing a UCI bug. That's good to know though. I think I'll just download polyglot so that I can have a "UCI GUI".

[Dann Corbit]

[d]rn1qkbnr/pp1b1ppp/8/1Bpp4/3P4/8/PPPNQPPP/R1B1K1NR b KQkq - bm Qe7; id "sbd.120";

This is a well known position in the french defense. The move of choice is Be7, not Qe7. Qe7 has been played but it is not the one preferred by the theory.

Miguel

I would be curious to know if Be7 really is better. I also tested the moves statistically for highest point percentage.

[Dann Corbit]

Here are some results by strong engines at 30 seconds per position (I would have more but the tests hung when it got to GarboChess):

Code: Select all

Analyzing engine: Alaric707            90 of 134 matching moves
Analyzing engine: AnMon_560           120 of 134 matching moves
Analyzing engine: Alfil811Optimized   114 of 134 matching moves
Analyzing engine: Arasanx 10.3         94 of 134 matching moves
Analyzing engine: Aristarch 4.50      113 of 134 matching moves
Analyzing engine: BaronUci            105 of 134 matching moves
Analyzing engine: Booot413             94 of 134 matching moves
Analyzing engine: Bright-0.3a         113 of 134 matching moves
Analyzing engine: ChessTiger2007UCI   119 of 134 matching moves
Analyzing engine: Colossus2007d       120 of 134 matching moves
Analyzing engine: Deep Frenzee        107 of 134 matching moves
Analyzing engine: Delfi               127 of 134 matching moves
Analyzing engine: DeepSjeng 1.6       117 of 134 matching moves
Analyzing engine: EngineShredder10UCI 114 of 134 matching moves
Analyzing engine: Fruit-2-3-1         122 of 134 matching moves
Analyzing engine: Gandalf60           128 of 134 matching moves

[Dann Corbit]

[d]rn1qkbnr/pp1b1ppp/8/1Bpp4/3P4/8/PPPNQPPP/R1B1K1NR b KQkq - bm Qe7; id "sbd.120";

This is a well known position in the french defense. The move of choice is Be7, not Qe7. Qe7 has been played but it is not the one preferred by the theory.

Miguel

I would be curious to know if Be7 really is better. I also tested the moves statistically for highest point percentage.

You are clearly right about this one. From my "high quality" database, I get 95% Be7 and a much better winning score than Qe7. Not sure how I goofed it up, but Be7 looks like the better move.

[gladius]

Here are some results by strong engines at 30 seconds per position (I would have more but the tests hung when it got to GarboChess):

Can you send me the command that you are using to analyze the positions? I probably have a UCI bug somewhere there, and it would be good to fix.

Thanks.

[Dann Corbit]

[d]rn1qkbnr/pp1b1ppp/8/1Bpp4/3P4/8/PPPNQPPP/R1B1K1NR b KQkq - bm Qe7; id "sbd.120";

This is a well known position in the french defense. The move of choice is Be7, not Qe7. Qe7 has been played but it is not the one preferred by the theory.

Miguel

I would be curious to know if Be7 really is better. I also tested the moves statistically for highest point percentage.

You are clearly right about this one. From my "high quality" database, I get 95% Be7 and a much better winning score than Qe7. Not sure how I goofed it up, but Be7 looks like the better move.

I am going to reverify all of the positions using the database tonight, since this one was so awfully done.

[Dann Corbit]

Here are some results by strong engines at 30 seconds per position (I would have more but the tests hung when it got to GarboChess):

Can you send me the command that you are using to analyze the positions? I probably have a UCI bug somewhere there, and it would be good to fix.

Thanks.

I was running under Arena 1.1, doing an EPD test set against a collection of engines.
If you have a debug setting that will save the information you need, I can rerun it for you.