Author: blass uri
Date: 23:50:06 04/14/00
Go up one level in this thread
On April 15, 2000 at 01:29:18, KarinsDad wrote:
>On April 15, 2000 at 00:08:41, blass uri wrote:
>
>>On April 14, 2000 at 14:41:08, KarinsDad wrote:
>>
>>>On April 13, 2000 at 20:32:01, Robert Hyatt wrote:
>>>
>>>[snip]
>>>>
>>>>You are talking about "forward pruning" which is full of danger. Your
>>>>eval likely doesn't understand deep tactics, yet you will be letting it
>>>>dictate which moves you search looking for tactics and which you don't.
>>>>
>>>>It is _very_ difficult to do this and not cause huge search problems...
>>>
>>>
>>>That is why Fathom will be so dangerous ("yeah, right" says all of the old
>>>programmers). It does a type of forward pruning. Whether it will be successful
>>>with this, only time will tell.
>>
>>I understand that you mean to selective search(this is what the old programmers
>>did) that and not to searching part of the moves to a reduced depth.
>>
>>
>>I think that the first thing to do in order to check these pruning ideas is to
>>take a database of about 10,000 comp-comp games(you can use the ssdf games) and
>>see if your pruning idea rejects the moves that were played.
>>
>>If you reject good moves that were played even only in one case out of 1000
>>positions then your forward pruning try is not good.
>>
>>If you only reject some bad moves that were played because of bugs then your
>>pruning idea may be good(probability not more than 1/1000000 for an error)
>>
>>Uri
>
>I agree that this is a place to start, but I think your numbers are way too
>restrictive. Many positions have 2 to 4 great moves, not just 1. So, if my
>program comes up with the comp chosen move 40% of the time or higher, I will be
>feeling pretty good.
Maybe I did not explain myself well.
The main problem is the error of prunning of good moves because it can cause
tactical mistakes and not the error of not prunning bad moves.
I do not expect your program to prune all moves except 2 to 4 great moves.
If you can prune only 20% of the moves because you are sure that they are bad
when the probability to be wrong in prunning is less than 1 to 1000000 then your
idea may be good.
If you prune good moves in 1 out of 1000 cases then you can lose by tactical
errors so I think the pruning idea is not good and the fact that you get bigger
depthes does not compensate for the tactical errors that may happen in more than
1 out of 1000 cases because a mistake of prunning not in the root may lead to a
wrong move.
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.