Author: Albert Silver
Date: 22:33:28 02/12/00
Go up one level in this thread
On February 12, 2000 at 21:03:07, stuart taylor wrote:
>On February 12, 2000 at 20:18:32, Albert Silver wrote:
>
>>On February 12, 2000 at 20:08:25, stuart taylor wrote:
>>
>>>On February 11, 2000 at 13:34:14, Albert Silver wrote:
>>>
>>>>On February 11, 2000 at 09:08:28, stuart taylor wrote:
>>>>
>>>>>On February 11, 2000 at 08:47:46, Albert Silver wrote:
>>>>>
>>>>>>On February 11, 2000 at 08:34:16, Vincent Vega wrote:
>>>>>>
>>>>>>>I understand that somebody is working on confirming whether there is a linear
>>>>>>>ELO increase with ply depth. If this indeed proves to be the case (as earlier
>>>>>>>results show), the slow searchers will get the same benefit with the increase of
>>>>>>>processor speed as the fast searchers. On the other hand, if there is a falloff
>>>>>>>somewhere, watch out for CSTAL, etc. They will rule in a couple years.
>>>>>>
>>>>>>I don't see how anyone could confirm a linear ELO increase with ply depth unless
>>>>>>a large amount of games were played with limited depths against human players.
>>>>>>Unless you are talking about computer-computer games where the effect is far
>>>>>>more decisive.
>>>>>>
>>>>>
>>>>>What's the rationale behind the possibility that greater depth may not
>>>>>necesarily result in better decisions over the board?
>>>>> Do you mean brute force vs. selective search, or great depth branching off a
>>>>>very limited ply count? (which is like very selective search with good pruning)
>>>>>S.Taylor
>>>>
>>>>My tactics are most certainly limited by the moves I choose to analyze and how
>>>>deeply I choose to analyze them, but that's just the tactics. My positional
>>>>play, my plans, my understanding of the position will not be changed because I
>>>>saw a ply deeper. If I realize that in position X an exchange of the queens and
>>>>one pair of rooks will result in a possibly won endgame, I don't see how seeing
>>>>even 10 extra plies will make up for that. That's knowledge, and a ply here or a
>>>>ply there won't outweigh it. Look at that famous position of Rebel where it had
>>>>its bishop locked in. Do you think that it will suddenly see the problem because
>>>>it is calculating a bit deeper?
>>>>
>>>
>>> I think that re. the famous position with Rebel and his bishop trapped in
>>>could possibly be different with a ply or two deeper. Of course it must be
>>>combined with knowledge to be applied to what it is seeing. But I beleive
>>>that other programs would avoid such things because it will sense the lack
>>>of power in the resulting position even without knowing the reason for it.
>>
>>Rebel is a computer program you know. I've never heard of a program feeling
>>anything, much less doing something without having been programmed for it. As
>>for the trapped bishop, I'm afraid not even 20 plies will do it. Don't take my
>>word for it though, try it. Run the next 10 moves once the bishop was trapped,
>>and see what Rebel Century says. Either the program can see it because the
>>knowledge is in it, or it can't.
>>
>> Albert Silver
>>
>>>I don't have the program, but I beleive you on that.
>I thought that a program CAN "feel" what it wasn't programed to actually
>understand.
No. It's a nice dream though. :-) You should take a look at the source code of
GNU chess or Crafty. You'll see that it has knowledge implemented each with a
value attributed to it, much like the piece values. So it will look at a
position and first calculate the material. Let's suppose it is equal so no
advantage for anyone there. Now for the positional knowledge:
Rook on an open file..............................+0.10
Rook on the 7th...................................+0.00 (no rooks on the 7th)
Doubled pawns.....................................-0.15
Isolated pawns....................................-0.05
Penalty for isolated pawn that is also doubled....-0.25
Etc.....
Total.............................................-0.35
And that will be the evaluation you will see.
> If it looked ahead 10-20 plies, wouldn't it realize that the other side
>had the upper hand atleast slightly-compared to if he did a move that would
>not close off the bishop?
No. How will the computer know the bishop can NEVER get out unless you show it
how to see this? If you limit it to the range of the piece, it will only see the
piece has a limited range, but unless you also show it how to see the cases
where the bishop will never get out and give it a negative value accordingly, it
will only see that it has a slightly limited range (much like any position).
You can't even give it a fatal value for a limited range as this would be wrong
most of the times, so the ONLY way is for either the program to see a forced
loss (which is what this should cause) which in the Rebel game was beyond the
range (though probably not the undersatnding) of even Deep Blue, or teach it how
to see this phenomenon.
> The evaluation function of rebel is of course at fault, but I don't think
>the actual knowledge need be the only thing to enable positionally wise
>moves to be made by a computer.
A ply can only reveal what the computer understands. See my discussion with VV
as I don't want to start a lot of cutting and pasting. (Plus it's pretty late
here).
Albert Silver
> (If you still don't agree,I wonder what Dr. Hyatt would say.)
> Stuart Taylor
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.