Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Nolot test fun!

Author: Robert Hyatt

Date: 11:24:16 01/13/00

Go up one level in this thread


On January 13, 2000 at 03:37:19, Dann Corbit wrote:

>On January 13, 2000 at 01:11:19, Jeremiah Penery wrote:
>[snip]
>>For #1, I don't think the key is finding Nxh6.  The key is to find that it is
>>winning, namely by finding 2. Nf5!  If they play Nxh6 with a -1 eval or so (as
>>all the programs on your page were doing), and follow up with inferior moves,
>>they are still going to lose.  The test is really about finding _winning_ moves,
>>and IMO, you must find that the move is winning to have really solved any of
>>these positions.
>
>This is a bone of contention with me as well.  If an eval shows -1000 and the
>right move is chosen, it means nothing if the eval should be +1000.  I think
>there should also be a minimal ce value for important tests to know if they
>really found the move.  It would also remove a loophole used by some programs to
>"exit early" as soon as they happen to hit the right move (for any reason --
>without regard to ce).  "LOOK HOW FAST WE SOLVE THE TEST!!"  Sure, but your
>program thought it was losing on 40% of the answers.
>:-(


I am not sure if you are talking about me, but when I post/publish test results,
I _never_ "early exit".  I put the early exit feature into my test code to make
debugging more pleasant.  IE if I hit and hold the key move for N plies, where
N can be specified, then I can stop and go on to the next one.  This turns a
1 minute per position win at chess run into a 3 minute deal, rather than a 300
minute deal.  Which lets me sanity check an extension modification quickly.

But for real test results, I _always_ search every position for the specified
amount of time to be sure that it doesn't change.

Bob



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.