During ACCA, CC told TT that Romi has many bugs

[Edsel Apostol]

My mistake.

I didn't look it up in my source code. The one you did is correct, just like what I'm doing.

Thanks Tony for clarifying it.

[Michael Sherwin]

It would have been nice if CC would tell me what they are.

Well, I think that I found one.

I do an aspiration search and when it fails, an 'Infinity' window is used for the research.

The bug would be to timeout and therefore not to have an accurate score for the iteration. How bad is this bug?


What should I do?

It can be very bad. I have played games where I would get a fail low ( phew, luckily my engine saw that) and then still have it play the crappy move (Oops, timeout) while it was still finding out, how crappy the move was.

Solution is on a faillow (fe >0.5 pawn), set the timeout to half the available time.

Other solution is to drop the aspiration window. It's not that good, most testing is flawed.

It will shorten the time for the first move, but no aspiration window ( with fail soft) is faster for the remaining moves. All in all, not much difference.

Tony

I have:

1.) allocated 1/2 the remaining time for failing low/high

2.) reduced the time seen to 80% of what's availible to conserve time

3.) using zero-window at the root (wanted to test just usage at the root first)

Sofar, Romi has won the first 6 games against Hamsters2 and DanaSah285 and is drawing the 7th.

[Christopher Conkie]

It would have been nice if CC would tell me what they are.

Well, I think that I found one.

I do an aspiration search and when it fails, an 'Infinity' window is used for the research.

The bug would be to timeout and therefore not to have an accurate score for the iteration. How bad is this bug?


What should I do?

It can be very bad. I have played games where I would get a fail low ( phew, luckily my engine saw that) and then still have it play the crappy move (Oops, timeout) while it was still finding out, how crappy the move was.

Solution is on a faillow (fe >0.5 pawn), set the timeout to half the available time.

Other solution is to drop the aspiration window. It's not that good, most testing is flawed.

It will shorten the time for the first move, but no aspiration window ( with fail soft) is faster for the remaining moves. All in all, not much difference.

Tony

I have:

1.) allocated 1/2 the remaining time for failing low/high

2.) reduced the time seen to 80% of what's availible to conserve time

3.) using zero-window at the root (wanted to test just usage at the root first)

Sofar, Romi has won the first 6 games against Hamsters2 and DanaSah285 and is drawing the 7th.

That sounds good Mike. In general Romi was losing alot of time in the ACCA games. I have not seen Romi do that before. Out of interest and completely unrelated, here is our recent OW game. It is odd how Romi does not see the score as 0.00 when the rep was in its pv.

[Event "OpenWar 4th Edition 30+5"]
[Site "DELL-E6600"]
[Date "2008.07.08"]
[Round "61.19"]
[White "Azrael 1.0m"]
[Black "RomiChess P3k"]
[Result "1/2-1/2"]
[TimeControl "1800+5"]
[Annotator "9. -0.37 2... +0.23"]
[Number "2119"]

1. d4 Nf6 2. c4 e6 {+0.23/10} 3. Nf3 d5 {-0.19/9} 4. Nc3 c6 {+0.08/9} 5.
Bg5 dxc4 6. e3 b5 {-0.30/17} 7. a4 Bb4 {+0.05/17} 8. Nd2 Bb7 {-0.03/17} 9.
Qf3 {-0.37/10} a6 {+0.06/17} 10. axb5 {-0.17/11} axb5 {-0.07/18} 11.
Rxa8 {-0.28/13} Bxa8 {+0.23/19} 12. Bxf6 {-0.38/13} gxf6 {+0.18/18} 13.
Be2 {-0.31/12} Nd7 {+0.04/16} 14. O-O {-0.34/11} c5 {-0.09/17} 15.
Qg4 {-0.23/11} cxd4 {-0.17/16} 16. Qxd4 {-0.28/10} Rg8 {-0.08/16} 17.
g3 {-0.31/10} Bxc3 {-0.34/16} 18. Qxc3 {+0.13/11} Qb8 {+0.07/15} 19.
b3 {+0.48/11} cxb3 {-0.50/15} 20. Bxb5 {+0.58/10} Qxb5 {-0.02/17} 21.
Qc8+ {+1.44/11} Ke7 {+0.24/6} 22. Qxg8 {+1.25/11} Bc6 {+0.17/16} 23.
Qg4 {+1.56/10} Ne5 {+0.65/17} 24. Qd4 {+0.00/11} Qe2 {+0.70/17} 25.
Qb4+ {+0.00/11} Ke8 {+0.74/18} 26. Qb8+ {+0.00/12} Ke7 {+0.74/19} 27.
Qb4+ {+0.00/12} Ke8 {-0.73/18} 28. Qb8+ {+0.00/13} Ke7 {-0.72/18} 29.
Qb4+ {+0.00/13}
{Draw by repetition} 1/2-1/2

[Michael Sherwin]

It would have been nice if CC would tell me what they are.

Well, I think that I found one.

I do an aspiration search and when it fails, an 'Infinity' window is used for the research.

The bug would be to timeout and therefore not to have an accurate score for the iteration. How bad is this bug?


What should I do?

It can be very bad. I have played games where I would get a fail low ( phew, luckily my engine saw that) and then still have it play the crappy move (Oops, timeout) while it was still finding out, how crappy the move was.

Solution is on a faillow (fe >0.5 pawn), set the timeout to half the available time.

Other solution is to drop the aspiration window. It's not that good, most testing is flawed.

It will shorten the time for the first move, but no aspiration window ( with fail soft) is faster for the remaining moves. All in all, not much difference.

Tony

I have:

1.) allocated 1/2 the remaining time for failing low/high

2.) reduced the time seen to 80% of what's availible to conserve time

3.) using zero-window at the root (wanted to test just usage at the root first)

Sofar, Romi has won the first 6 games against Hamsters2 and DanaSah285 and is drawing the 7th.

That sounds good Mike. In general Romi was losing alot of time in the ACCA games. I have not seen Romi do that before. Out of interest and completely unrelated, here is our recent OW game. It is odd how Romi does not see the score as 0.00 when the rep was in its pv.

[Event "OpenWar 4th Edition 30+5"]
[Site "DELL-E6600"]
[Date "2008.07.08"]
[Round "61.19"]
[White "Azrael 1.0m"]
[Black "RomiChess P3k"]
[Result "1/2-1/2"]
[TimeControl "1800+5"]
[Annotator "9. -0.37 2... +0.23"]
[Number "2119"]

1. d4 Nf6 2. c4 e6 {+0.23/10} 3. Nf3 d5 {-0.19/9} 4. Nc3 c6 {+0.08/9} 5.
Bg5 dxc4 6. e3 b5 {-0.30/17} 7. a4 Bb4 {+0.05/17} 8. Nd2 Bb7 {-0.03/17} 9.
Qf3 {-0.37/10} a6 {+0.06/17} 10. axb5 {-0.17/11} axb5 {-0.07/18} 11.
Rxa8 {-0.28/13} Bxa8 {+0.23/19} 12. Bxf6 {-0.38/13} gxf6 {+0.18/18} 13.
Be2 {-0.31/12} Nd7 {+0.04/16} 14. O-O {-0.34/11} c5 {-0.09/17} 15.
Qg4 {-0.23/11} cxd4 {-0.17/16} 16. Qxd4 {-0.28/10} Rg8 {-0.08/16} 17.
g3 {-0.31/10} Bxc3 {-0.34/16} 18. Qxc3 {+0.13/11} Qb8 {+0.07/15} 19.
b3 {+0.48/11} cxb3 {-0.50/15} 20. Bxb5 {+0.58/10} Qxb5 {-0.02/17} 21.
Qc8+ {+1.44/11} Ke7 {+0.24/6} 22. Qxg8 {+1.25/11} Bc6 {+0.17/16} 23.
Qg4 {+1.56/10} Ne5 {+0.65/17} 24. Qd4 {+0.00/11} Qe2 {+0.70/17} 25.
Qb4+ {+0.00/11} Ke8 {+0.74/18} 26. Qb8+ {+0.00/12} Ke7 {+0.74/19} 27.
Qb4+ {+0.00/12} Ke8 {-0.73/18} 28. Qb8+ {+0.00/13} Ke7 {-0.72/18} 29.
Qb4+ {+0.00/13}
{Draw by repetition} 1/2-1/2

Romi has a very high negitive contempt factor for a draw that zero's out at ply 100 and then becomes positive after that. Romi seen the draw at move 27 and scored it at -0.73 (100-27=73) at move 28 the draw became -0.72. Romi allowed the repitition because she could find nothing better. The question is then, was there something better.

You seem to have made progress with Azrael!

[jdart]

As others have noted it is fairly common to set an aspiration window at the root and re-search if it is exceeded.

This is similar to but independent of using zero-window for non-PV nodes, i.e. setting beta to alpha+1 after the first move has been searched. The difference is the aspiration window narrows the score bounds even for the first (PV) move.

It should also be mentioned that it is common to extend the search time if there is a fail-low at the root - i.e. an iteration has completed and the lower bound has not been exceeded, so it is necessary to re-search with a window bound of -Infinity (or a lower but not infinite bound if you are doing 2-stage widening).

Finally I might mention that as far as I can tell (but I could be wrong) Toga, and also other recent programs I have seen, uses neither an aspiration window nor zero-window searches. Toga does other unusual things, so it is not clear how this factor alone affects its search.

[Michael Sherwin]

As others have noted it is fairly common to set an aspiration window at the root and re-search if it is exceeded.

This is similar to but independent of using zero-window for non-PV nodes, i.e. setting beta to alpha+1 after the first move has been searched. The difference is the aspiration window narrows the score bounds even for the first (PV) move.

It should also be mentioned that it is common to extend the search time if there is a fail-low at the root - i.e. an iteration has completed and the lower bound has not been exceeded, so it is necessary to re-search with a window bound of -Infinity (or a lower but not infinite bound if you are doing 2-stage widening).

Finally I might mention that as far as I can tell (but I could be wrong) Toga, and also other recent programs I have seen, uses neither an aspiration window nor zero-window searches. Toga does other unusual things, so it is not clear how this factor alone affects its search.

Thanks Jon,

I am learning that there are alot of illusions when it comes to chess programming. I think that many bad ideas have been propagated over the years just because it is taken for granted that they are good if they are being used by a strong program. In my opinion some 'good' ideas are simply left in if it is determined that they do not hurt. They may even give a small plus during testing. Later, however, they may limit a programs strength.