Houdini 1.03a bug-fix is available

[LetoAtreides82]

I've been running a gauntlet for Houdini 1.03 x64 4CPU with my normal CEGT blitz testing procedures, and after 144 out of 500 games the current scores are as follows:

vs Rybka 4 x64 4CPU (3258) 10.5-4.5 70% !!
vs Rybka 3 x64 4CPU (3235) 9-6 60%
vs Stockfish 1.8 x64 4CPU (????) 7.5-7.5 50% ?
vs Stockfish 1.7.1 x64 4CPU (3205) 7.5-6.5 53.57%
vs Naum 4.2 x64 4CPU (3158) 10-5 66.67%
vs Naum 4.1 x64 4CPU (3137) 9.5-4.5 67.86%
vs Naum 4 x64 4CPU (3121) 8-6 57.14%
vs Deep Shredder 12 4CPU (3093) 10-4 71.43%
vs Hiarcs 13.1 4CPU (3089) 8.5-5.5 60.71%
vs Deep Fritz 11 4CPU (3066) 8.5-5.5 60.71%

If Houdini 1.03a keeps this up it will be the unofficial CEGT Blitz leader. It's interesting to note how well it performs against Rybka, and yet it struggles a bit with Stockfish.

[Robert Houdart]

Thanks for sharing the results of the tournament!
I wouldn't draw too many conclusions from just 15 games against any opponent; many times have I seen results like 11-4 in 15 games between two engines of identical strength. This apparently also happens in long TC matches, see the current Rybka-Stockfish match in which Rybka is on a 11.5-3.5 streak (+8 =7 !). Imagine that this had happened in the first 15 games of the match...

Robert

[Darkanyons]

Good day everyone, and specially to Robert. Just this hour, i encountered an error on Houdini 1.03a. I was again trying to defeat the chessmaster using this engine where in the end game i got this error: UCI implementation fault: illegal end of ponder search.

Shedder 11 could no long continue. I have to shut it down using taskmanager. have anyone encountered this? and most importantly is there solution for this?

Many thanks.

[Robert Houdart]

With ponder on, sometimes Houdini reaches search depth 100 while the opponent is still thinking, and then stops (as this is the maximum search depth). This generates the "illegal end of ponder search" in Shredder (not sure why this would crash the GUI, though).
It's basically a problem of Houdini thinking too fast, I'll try to avoid this in the next release by adding an empty loop until the GUI tells it to stop thinking.

I've only seen this in completely drawn pawn endgames, can you confirm this was also your case?

Cheers,
Robert

[LetoAtreides82]

I've been running a gauntlet for Houdini 1.03 x64 4CPU with my normal CEGT blitz testing procedures, and after 144 out of 500 games the current scores are as follows:

vs Rybka 4 x64 4CPU (3258) 10.5-4.5 70% !!
vs Rybka 3 x64 4CPU (3235) 9-6 60%
vs Stockfish 1.8 x64 4CPU (????) 7.5-7.5 50% ?
vs Stockfish 1.7.1 x64 4CPU (3205) 7.5-6.5 53.57%
vs Naum 4.2 x64 4CPU (3158) 10-5 66.67%
vs Naum 4.1 x64 4CPU (3137) 9.5-4.5 67.86%
vs Naum 4 x64 4CPU (3121) 8-6 57.14%
vs Deep Shredder 12 4CPU (3093) 10-4 71.43%
vs Hiarcs 13.1 4CPU (3089) 8.5-5.5 60.71%
vs Deep Fritz 11 4CPU (3066) 8.5-5.5 60.71%

If Houdini 1.03a keeps this up it will be the unofficial CEGT Blitz leader. It's interesting to note how well it performs against Rybka, and yet it struggles a bit with Stockfish.

After 260 out of 500 games, here are the results:

vs Rybka 4 x64 4CPU (3258) 16-10 61.54%
vs Rybka 3 x64 4CPU (3235) 14.5-11.5 55.7%
vs Stockfish 1.8 x64 4CPU (????) 12.5-13.5 48%
vs Stockfish 1.7.1 x64 4CPU (3205) 15-11 57.69%
vs Naum 4.2 x64 4CPU (3158) 180.5-7.5 71.15%
vs Naum 4.1 x64 4CPU (3137) 17-9 65.38%
vs Naum 4 x64 4CPU (3121) 16.5-9.5 63.46%
vs Deep Shredder 12 4CPU (3093) 18-8 69.23%
vs Hiarcs 13.1 4CPU (3089) 16-10 61.54%
vs Deep Fritz 11 4CPU (3066) 17.5-8.5 67.31%

Houdini 1.03a looking good so far except for its match vs Stockfish 1.8.

[Darkanyons]

I've only seen this in completely drawn pawn endgames, can you confirm this was also your case?

Cheers,
Robert

Thank you for the explanation.

Since i am quite new on chess, i honestly don't know if this is a drawn pawn endgame. however, perhaps a picture of the scenario may help:


the last move was white to h3. Black was played by houdini 1.03a engine. this is where i got the error.

[Robert Houdart]

No, this is not a draw pawn ending, in this position Black mates in 3 moves: 1...Bxe2+ 2.Kg1 Bc5+ 3.Kh2 Qxf4#.
The situation is similar to what I described: Houdini very rapidly reaches the maximum search depth, and then has nothing more to ponder about. I'll try to implement a solution for this in the next release.

Thanks for reporting this,
Robert

[Darkanyons]

Your welcome.

And thank you so much for explaining and attending this issue. I'm eagerly looking forward to any future updates on Houdini in your due time.

[LetoAtreides82]

I've been running a gauntlet for Houdini 1.03 x64 4CPU with my normal CEGT blitz testing procedures, and after 144 out of 500 games the current scores are as follows:

vs Rybka 4 x64 4CPU (3258) 10.5-4.5 70% !!
vs Rybka 3 x64 4CPU (3235) 9-6 60%
vs Stockfish 1.8 x64 4CPU (????) 7.5-7.5 50% ?
vs Stockfish 1.7.1 x64 4CPU (3205) 7.5-6.5 53.57%
vs Naum 4.2 x64 4CPU (3158) 10-5 66.67%
vs Naum 4.1 x64 4CPU (3137) 9.5-4.5 67.86%
vs Naum 4 x64 4CPU (3121) 8-6 57.14%
vs Deep Shredder 12 4CPU (3093) 10-4 71.43%
vs Hiarcs 13.1 4CPU (3089) 8.5-5.5 60.71%
vs Deep Fritz 11 4CPU (3066) 8.5-5.5 60.71%

If Houdini 1.03a keeps this up it will be the unofficial CEGT Blitz leader. It's interesting to note how well it performs against Rybka, and yet it struggles a bit with Stockfish.

After 260 out of 500 games, here are the results:

vs Rybka 4 x64 4CPU (3258) 16-10 61.54%
vs Rybka 3 x64 4CPU (3235) 14.5-11.5 55.7%
vs Stockfish 1.8 x64 4CPU (????) 12.5-13.5 48%
vs Stockfish 1.7.1 x64 4CPU (3205) 15-11 57.69%
vs Naum 4.2 x64 4CPU (3158) 180.5-7.5 71.15%
vs Naum 4.1 x64 4CPU (3137) 17-9 65.38%
vs Naum 4 x64 4CPU (3121) 16.5-9.5 63.46%
vs Deep Shredder 12 4CPU (3093) 18-8 69.23%
vs Hiarcs 13.1 4CPU (3089) 16-10 61.54%
vs Deep Fritz 11 4CPU (3066) 17.5-8.5 67.31%

Houdini 1.03a looking good so far except for its match vs Stockfish 1.8.

Gauntlet completed here are the results:

vs Rybka 4 x64 4CPU (3258) 26.5-23.5 53%
vs Rybka 3 x64 4CPU (3235) 28.5-21.5 57%
vs Stockfish 1.8 x64 4CPU (3196?) 25.5-24.5 51%
vs Stockfish 1.7.1 x64 4CPU (3205) 29.5-20.5 59%
vs Naum 4.2 x64 4CPU (3158) 33-17 66%
vs Naum 4.1 x64 4CPU (3137) 31.5-18.5 63%
vs Naum 4 x64 4CPU (3121) 32.5-17.5 65%
vs Deep Shredder 12 4CPU (3093) 38-12 76%
vs Hiarcs 13.1 4CPU (3089) 30-20 60%
vs Deep Fritz 11 4CPU (3066) 34.5-15.5 69%

With plus scores against all opponents it is safe to say Houdini 1.03a is the new unofficial king of blitz. I'll try to get an elo this sunday, unfortunately i'm too busy at the moment to do it.

[snino64]

I am looking at various positions with Houdini 1.03a multipv mode = 4.
If I pause the engine just as you are thinking about number one variation, the engine fails to send a program change number (SCID), I can only observe the variations whose numbers are 2 3 and 4.
For example, in this position is mate in 19 moves with NF7, but the engine just pause when thinking about the first play (NF7), this variation disappears from the engine information and lines that appear are:
1 32 M21 1.Be4 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6#
2 32 M21 1.Bd3 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6#
3 32 M21 1.Bf5 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6#

[Event "?"]
[Site "?"]
[Date "????.??.??"]
[Round "?"]
[White "?"]
[Black "?"]
[Result "*"]
[FEN "6k1/8/5KB1/6N1/8/8/8/8 w - - 0 1"]

1.Nf7
( [Houdini 1.03a x64 8_CPU] 32:M21 1.Be4 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6# )

( [Houdini 1.03a x64 8_CPU] 32:M21 1.Bd3 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6# )

( [Houdini 1.03a x64 8_CPU] 32:M21 1.Bf5 Kh8 2.Nf7+ Kg8 3.Bg6 Kf8 4.Bh7 Ke8 5.Ne5 Kd8 6.Ke6 Kc7 7.Nd7 Kb7 8.Bd3 Kc6 9.Bc4 Kc7 10.Bb5 Kd8 11.Nb6 Kc7 12.Na4 Kd8 13.Kd6 Kc8 14.Nc5 Kd8 15.Ne6+ Kc8 16.Ba6+ Kb8 17.Kc6 Ka8 18.Kb6 Kb8 19.Nc5 Ka8 20.Bb7+ Kb8 21.Na6# )