Author: Uri Blass
Date: 18:22:21 11/20/01
Go up one level in this thread
On November 20, 2001 at 14:54:34, Gordon Rattray wrote: >On November 20, 2001 at 14:34:40, Uri Blass wrote: > >>On November 20, 2001 at 12:33:15, Gordon Rattray wrote: >> >>>On November 20, 2001 at 11:05:36, Martin Giepmans wrote: >>> >>>[snip] >>> >>>>I like the learning-feature of Yace. >>> >>>Sorry, I've never used Yace... what is the "learning feature"? Is it similar to >>>some other programs where they can update their hash tables and then have moves >>>retracted? >>> >>>Gordon >> >>similiar but better than most programs. >> >>If I analyze with Junior line A go back and later analyze line B and go back >>then it does not forget what it learned from line A. >> >>My experience with part of the other programs is different and if I analyze line >>B for enough time they can forget what they learned from line A. >> >>In this case it was not relevant because there was no line B that I needed to >>analyze for a long time but there are cases when it is important. >> >>Uri > > >I assume you do this because you believe that it may help find a better move as >opposed to solely searching from the root position? If so, aren't you making >the overall search deeper but also more selective? And hence your own chess >judgement is important? The score of the program is not based on selective search. The only difference between it and regular search is that it remembers the score of specific positions and use them instead of searching to reduced depth at them. Example: In the opening position if it analyzed the position after 1.e4 e5 2.Nf3 to depth 13 and return to the root and get remaining depth 11 for the position after 1.e4 e5 2.Nf3 then it does not need to analyze it because it trust the score at depth 13 more than the score at depth 11. > >Looked at another way, if "stepping in lines and then retracting" can give a >better root move, why doesn't the search from the root do this automatically? The problem is also to do it only in the right positions and it can be counter productive in quiet positions when the target is to find a better positional move. I believe that by doing it for a small part of the time it is possible to get at long time control tactical improvement when the positional problems from being 10% slower in most positions are going to be small but programmers usually do not care about long time control. >Maybe it's the user's chess skill that is a key factor?! It is also the user skill to choose the right lines but it does not mean that the analysis is wrong. In this case choosing the right line was based on previous analysis and it did the task easier. 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.