Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: an evaluation problem of chess programs

Author: blass uri

Date: 10:37:09 10/04/98

Go up one level in this thread



On October 04, 1998 at 13:06:30, Robert Hyatt wrote:

>On October 04, 1998 at 11:51:20, blass uri wrote:
>
>>
>>On October 04, 1998 at 10:00:09, Robert Hyatt wrote:
>>
>>>On October 04, 1998 at 05:24:42, blass uri wrote:
>>>
>>>>
>>>>On October 03, 1998 at 22:07:23, Robert Hyatt wrote:
>>>>
>>>> >It really wouldn't help, because this would reduce the size of the current
>>>>>tablebases by 1/4, because they use 8 bit values.
>>>>
>>>>How can you use 8 bit values
>>>>only to store the move you may need 6 bits because you may have more than 32
>>>>legal moves even only with king and queen.
>>>>
>>>>do you mean 8 bits only to store the result and that the tablebases use more
>>>>than 8 bits for position(8 bits for result and 5 or 6 bits for the move)?
>>>>
>>>>you can save space in the harddisk by not storing the result and computing it by
>>>>playing the right moves against yourself but in this case you are slower.
>>>>
>>>>Uri
>>>
>>>
>>>
>>>there are no "moves" in the database.  It is a huge stream of bytes indexed by
>>>a "Godel" number that is simply a concatenation of the squares each piece sit
>>>on.  IE let's number the squares a1=0...  h8=63.  put the white knight on a1,
>>>white king on b1, white bishop on c1 and black king on d1.  All we do is take
>>>the 4 squares and compute b1<<18 + a1<<12 + c1<<6 + d1...  which gives us a
>>>number that indexes into the database, and the value we find there is +Mate in
>>>N, -Mate in N, or Draw (0).
>>>
>>>No moves, no pieces, etc.  The pieces (the squares they stand on) form the
>>>key, the result at that position is how many moves to mate, or else "draw"
>>
>>I understand
>>You can  use the exact result of mate in N to compute the move.
>>
>>I think that 7 bytes may be enough to store the result if instead of storing
>>mate in how many moves you store only the number of moves before the next
>>capture or the next moving of a pawn
>>because of the 50 moves rule you have only 101 possible results
>>draw or win in 1-50 moves for you or win in 1-50 moves for the opponent
>>
>>Uri
>
>
>I assume you mean "seven bits".  But this doesn't tell me anything about how
>to choose between two moves in the tree, when one is mate in 10 and one is
>mate in 30.  And if I keep choosing mate in 30, I encounter problems, because
>I can end up drawing by the 50 move rule, since the "distance to conversion"
>doesn't factor in the moves played *before* this position was reached, only
>what happens *after* this position was reached...
>
>In any case, distance to mate is trivial to use, and never screws up.

theoretically distance to mate may cause problems becuase if one is mate in 70
that is practically a draw because of the 50 move rule and the second is mate in
80 that is not a draw because there is a capture after 40 moves than mate in 80
is the right move.

I do not understand how distance to the next capture or moving a pawn may cause
problems because what happened before is not relevant

> I can
>show you cases where fritz lost in drawn positions because it took the move
>that led to the "most distant conversion" and overlooked a mate in 2 on the
>board...

I want to see

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.