Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: test suites solicitation

Author: Dann Corbit

Date: 18:38:12 01/20/06

Go up one level in this thread


On January 20, 2006 at 21:34:45, Stuart Cracraft wrote:

>On January 20, 2006 at 01:46:06, Dann Corbit wrote:
>
>>On January 19, 2006 at 23:49:59, Stuart Cracraft wrote:
>>
>>>On January 19, 2006 at 13:47:05, Dann Corbit wrote:
>>>
>>>>On January 19, 2006 at 12:23:51, Stuart Cracraft wrote:
>>>>
>>>>>Hi - I have reached the limit of what I can test with Fred Reinfeld's
>>>>>Win-at-Chess suite of 300 positions. Not that I couldn't improve more
>>>>>points on it - but that many retests reach a plateau that does not
>>>>>change much.
>>>>>
>>>>>So, I want to expand to these suites:
>>>>>
>>>>>   WAC
>>>>>   1001 Winning Chess Sacrifices
>>>>>   1001 Brilliant Ways to Mate
>>>>>   ECO Middlegames
>>>>>   ECO Endgames
>>>>>
>>>>>and any others you think reasonable and prudent. My goal is a single
>>>>>overnight 8 hour test at a few seconds per position - folded into
>>>>>a single result with per-suite specifics as well.
>>>>>
>>>>>My issue is where to get the above in EPD format with the solution(s) included
>>>>>in each EPD entry.
>>>>>
>>>>>I think with the larger set plus my current adding of chess knowledge to
>>>>>terminal nodes, I will be able to make new progress in new areas and
>>>>>not be "WAC-blind" to it.
>>>>>
>>>>>Thanks ahead if you know of where I can get the above or other (better)
>>>>>suites.
>>>>
>>>>Most test suites are full of bugs.
>>>>Yuriy Lyapko has a bunch of carefully debugged ones.
>>>>You might ask him about it.
>>>>http://www.geocities.com/lyapko/index.html
>>>
>>>I notice you have an extremely large selection of test suites at
>>>
>>>http://cap.connx.com/EPD/
>>>
>>>Are you saying that none of these has relatively accurate solutions
>>>so as to be usable in chess program testing?
>>
>>I do not know the accuracy of most of the tests.  I have spent some time
>>verifying some of the easy ones like WAC, but they are of limited usefulness.
>>
>>For any of the test sets, I expect at least 80% of the answers to be valid.
>>
>>ECM-GCP is verified fairly well.  You could consider it as an alternative.
>
>Thanks - it's now test #2 for TeeDee - ran with it at 1 and 5 seconds. Looks
>much more challenging. What does the GCP stand for?

Gian-Carlo Pascutto




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.