Author: Robert Hyatt
Date: 09:04:23 03/04/99
Go up one level in this thread
On March 04, 1999 at 00:13:31, Dan Newman wrote:
>On March 02, 1999 at 23:09:26, Robert Hyatt wrote:
>
>>In case you didn't see my note on r.g.c.c, for those of you using win at chess
>>still, crafty found another 'cook' today. I was running this test just as a
>>'sanity test' on new changes, and at 30 secs/position, normally I get 299/300
>>correct. But in this run, I got 298. Investigation showed that it was now
>>missing 146, where the correct move is Bc8. A look at an old log showed
>>Bc8 very quickly with a score of +9, but a look at the new log showed Bd3 with
>>a score of +20. I let it run a bit longer and Bd3 leads to a mate in 51 via
>>tablebases. So there are two 'winning' moves. This 'change' came about by
>>adding all the new tablebase files from Eugene.
>>
>>Interesting stuff..
>>
>>
>>here is the output:
>>
>>log.003: 15 1:56 Mat51 1. Bd3 f2 2. c5 Ke3 3. cxd6 Kxd3
>><EGTB>
>>log.003: 15-> 2:08 Mat51 1. Bd3 f2 2. c5 Ke3 3. cxd6 Kxd3
>><EGTB>
>>
>>
>>I'd claim any move leading to a forced mate is an adequate solution. :)
>>
>>Bob
>
>Hey, neat! Now the question is, should I add this to my copy of WAC...
>I just tried mine out on 146, and it likes Bd3 almost from the start and
>holds it for 50 s till it fails high on Bc8. Of course it has no idea
>that Bd3 is any good -- just the least of several evils.
>
>-Dan.
This is one of the vague points. IE Reinfeld probably assumed that any move
other than Bc8 is bad. And unless you can 'see' why Bd3 works, it probably
should be considered 'bad' also. But in the case of test positions, we don't
generally require that the program see the 'point' only the 'move' which means
we quite often have 'cooks'.
1. finding the right move for the wrong reason is not too uncommon.
2. find the right move for the right reason positionally, but not seeing
the tactical 'truth' (ie the way I solve wac 2 at shallow depths, before
I see the win of material at deeper depths).
3. finding the right move with the right PV and score.
(3) is what we should all 'want'. But we end up accepting a few 1's and 2's
here and there as well.
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.