Author: Robert Hyatt
Date: 18:36:23 01/19/04
Go up one level in this thread
On January 19, 2004 at 18:57:12, Dieter Buerssner wrote:
>On January 19, 2004 at 17:51:27, Robert Hyatt wrote:
>
>>Someone else pointed out that it is possible that the move was a fail-high
>>or fail-low move. Printing out a PV on such moves is impossible and you get
>>_lots_ of garbage most of the time.
>
>I got the feeling, that printing out "PVs" (they are no "real" PVs, as we both
>know, but may be called "refutation lines") is surprisingly often not that full
>of garbage, as one might expect (when using fail soft alpha-beta). I would
>certainly not call it impossible. It gives more info (which one can ignore, and
>is back to the same, when not printing it, so no real disadvantage).
>
If you fail high at the root, you search all moves at ply=2 with no best
move found. What best move do you display for there? The move you will show
is a random move. If you do as I do and try to preserve the best move during
replacement, when it makes sense, then that is what you might show. And it
may or may not "fit" here. IE it might be an old move from a different window
search or whatever...
If you use true fail-soft, you _might_ get a decent move, but rarely the
best, because at ply-3 you don't always search the best move, just a move
good enough to produce a fail high, which will make trying to choose the
best move at ply=2 impossible. Serendipity might produce a good one, but
it will also produce bad ones as well.
>I just tried the Nf5 pos posted here recently.
>
> 91388636 2:08.1 -0.50 11t 16.Bxb7 Qxb7 17.Na5 Qd5 18.Nc4 Bd8H 19.Nxb3H
> Qxd1H 20.Rfxd1H O-OH 21.Nd6H {-80}
> 145628113 3:24.9 -0.49 11t+ 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bd6
>[Outside of window, probably now the line continues with "null", which my engine
>will not display by default. For debugging even displaying that and remaining
>moves can be interesting, to see at least a sample line that failed high.]
> 152281645 3:38.1 0.01 11++ 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bxf1 19.Nd6+
> Bxd6 20.Qxd6 Nc6 21.Rxc6 b2 22.Qe5+ Kf8H
> 23.Bc5+H {HT} {1}
>[alternative would be to just print Nf5!!]
> 170526494 4:06.5 0.29 11t 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bxf1 19.Nd6+
> Bxd6 20.Qxd6 Nc6 21.Rxc6 Ne4 22.Qe5+ Kf8 23.Rc7
> {1}
> 180964542 4:19.9 0.29 11. 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bxf1 19.Nd6+
> Bxd6 20.Qxd6 Nc6 21.Rxc6 Ne4 22.Qe5+ Kf8 23.Rc7
> {1}
> 223379378 5:30.2 0.69 12++ 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bxf1 19.Nd6+
> Bxd6 20.Qxd6 Nc6 21.Rxc6 Nd5 22.Bc5 Rd8 23.Qe5+
> Ne7 24.Kxf1 dxc6 25.Qxe7# {191}
>[And here, too]
> 247811846 6:11.8 0.94 12t 16.Nf5 exf5 17.Bd4 Bxg2 18.Bxa7 Bxf1 19.Nd6+
> Bxd6 20.Qxd6 Nc6 21.Rxc6 Ne4 22.Qe5+ Kf8 23.Rc7
> Bc4 24.Bc5+ Kg8 {HT} {1}
>
>At least the start of the lines, seems to look very reasonable (while it is all
>three times where noted, outside of the window). I also measured the overhead
>some time ago. Even when updating such "PVs" when outside of the window, the
>overhead seems totally neglible to me (and it will be updated then many more
>times, than in the case where updating it only when the score is inside the
>window).
>
>Regards,
>Dieter
I think the problem above is that when you run out of time after a fail high,
you have some bad moves showing that leave the wrong impression. IE move 25
above, where you end in mate but the score is not mate...
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.