Author: Uri Blass
Date: 21:06:24 09/21/04
Go up one level in this thread
On September 21, 2004 at 23:04:22, Robert Hyatt wrote:
>On September 21, 2004 at 23:01:38, Uri Blass wrote:
>
>>On September 21, 2004 at 22:44:33, Robert Hyatt wrote:
>>
>>>On September 21, 2004 at 21:53:47, Dann Corbit wrote:
>>>
>>>>On September 21, 2004 at 19:41:42, Stuart Cracraft wrote:
>>>>[snip]
>>>>>// white knight
>>>>> {
>>>>> OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF,
>>>>> OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF,
>>>>> OFF, 64, 41, 60, 67, 67, 60, 41, 64, OFF,
>>>>> OFF, 48, 79, 91, 76, 76, 91, 79, 48, OFF,
>>>>> OFF, 57, 86, 95, 111, 111, 95, 86, 57, OFF,
>>>>> OFF, 77, 105, 123, 129, 129, 123, 105, 77, OFF,
>>>>> OFF, 99, 124, 129, 145, 145, 129, 124, 99, OFF,
>>>>> OFF , 79, 114, 116, 144, 144, 116, 114, 79, OFF,
>>>>> OFF, 53, 77, 108, 81, 81, 108, 77, 53, OFF,
>>>>> OFF, 0, 71, 73, 73, 73, 73, 71, 0, OFF,
>>>>> OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF,
>>>>> OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF, OFF
>>>>> },
>>>>
>>>>A value of 64 for A8 and H8 seems highly suspect to me. Similarly for A1 and H1
>>>>for the black knight.
>>>
>>>
>>>Any positive value for _any_ square on the edge of the board seems to be more
>>>than "a bit suspect". They look outright _wrong_...
>>
>>values are relative so I see no problem with non negative values
>>I have only positive values in my piece square table.
>>
>>Everything that worth more than nothing is positive for me.
>>I have not one value for the knight but 64 values based on the square of the
>>knight.
>>
>>I see no reason to have special value for the knight except 64 values when 64
>>values already give all the picture.
>>
>>If I want to evaluate black knight at c3 then
>>pcsq[KNIGHT][c3] give me the relevant information and if you want to evaluate
>>white knight at c3 then pscq[KNIGHT][flip[c3]] can give you the same
>>information.
>>
>A knight on two of the corner squares is worth 0. On the other two 64. That
>doesn't make any sense to me...
I did not claim that his numbers are good but only that non negative numbers do
not have to be bad.
Uri
>
>
>>
>>>
>>>The problem is that it is very likely that the pc/sq values are not independent
>>>variables, they are closely associated with other eval terms modified during the
>>>tuning. Trying to transplant them to a different program/evaluation is a
>>>mistake...
>>
>>I agree about it.
>>
>>Uri
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.