Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Tactical speed and test suites

Author: Bruce Moreland

Date: 08:57:36 06/09/98

Go up one level in this thread



On June 08, 1998 at 21:45:43, Nobuhiro Yoshimura wrote:

>When you modify your search engine or evaluation functions and test
>aginst
>many test suites,  can you tell me how to decide whether you are going
>to
>adpot or reject the modification.
>
>For example:
>  before)   5sec  90% correct  and  10sec  95%correct
>    after)   5sec  85% correct  and  10sec  98%correct

I haven't had to make this decision.  If I had this situation, I'd look
at things more carefully to try to figure out why.  I'd also run other
suites.

90% is 270, 95% is 285.
85% is 255, 98% is 294.

255 is especially bad, but the difference between 294 and 285 is also
significant, although that'd be a place to look, to see if those extra
nine correct answers were gotten for the wrong reasons.

It would also be interesting to see how long it takes the first one to
get 294.

I would be inclined toward the second one, because hardware can only get
faster, but it might be interesting to see if the initial zip in the
first one can be incorporated into the second one.

You don't usually see such a dramatic difference, and when you do, there
isn't usually a tradeoff.  One will usually be obviously a little better
than the other.

bruce



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.