Author: Robert Hyatt
Date: 18:14:22 11/24/98
Go up one level in this thread
On November 24, 1998 at 20:37:36, Vincent Diepeveen wrote: >On November 24, 1998 at 08:12:46, Robert Hyatt wrote: > > >>>So there are 4 ways to test >>> a) all extensions off, SE on >>> b) all extensions on, SE on >>> c) all extensions on, SE off >>> d) all extensions off, SE off (which is the most uninteresting) >>> >>>The most interesting is to compare b with c. In fact S.E. should find this >>>really quickly, but when i test this with S.E. on, then i'm suffering horrible >>>from the reduction factor, which misses the advance of the c-pawn. >>> >>>In fact i find this move easier then without S.E. than with, as it eats >>>up too many plies extending nonsense. >>> >>>here is the position (bs2830-26): >>> >>>1r4k1/1q2pN1p/3pPnp1/8/2pQ4/P5PP/5P2/3R2K1 b - - Qd5 >> >> >>Did you implement the "real" singular extension algorithm as explained in the >>JICCA along with the "sticky transposition table" stuff to avoid losing a >>singularity at odd times? I did this in Cray Blitz and found it worked pretty >>well in the right positions... > >As described yes, i used a translated version of it to Netherlands. >If i remember well translated by Dennis Breuker in 'computerschaak'. here's why I asked... when I added this to Cray Blitz, it was a couple of thousand lines of code scattered around.. from the sticky transposition table, to all the exceptions and so forth. That's a bunch of code and when you have mentioned this in the past, you didn't give me the impression that this was a complicated thing to do. In reality it is quite complicated to eliminate most of the inconsistencies that crop up on each new iteration. > >Main problem is this reduction factor. >you first reduce in depth in order to see whether it's singular in order >to extend it. > >So you reduce it by 2 then extend it by 1. Of course: you reduce the >OTHER moves by 2, and extend the RIGHT move by 1, but how can you >DETECT that this right move is a singular move if you search the other >moves by a reduced depth, where the only interesting singularism is derived >from a search which is NOT reduced. the idea is "obviously singular"... and a 2 ply reduction doesn't hide an "obviously singular" move. It certainly will overlook "subtly singular" moves of course.. But it is similar in idea to the null-move search, which also works well as most of us know.. > >Further a human doesn't use 'singularism'. Most combinations you have 2 or >3 interesting moves, and not only 1 forced moves. singularism is just >a small subset of what humans do. Humans simply consider *all moves* >that are interesting. this is very sometimes 1 move, more usually 2 or 3 >or 4 moves. Sometimes a human simply 'nullmoves' or something like that. >scanning for potential winners for the opponent, using *a lot* of >chessknowledge to select a move. Not something with alfa or beta. > >Simply a line that *might* give counterplay without concerning about the >return value of a small search. > >Anyway, i think Singular Extensions is something interesting for old programs >with bad branching factors searching fullwidth. better watch out for those "old programs". The two I know of that use this will hand you your head in a covered basket... :) > >I think the future is for nullmove programs searching huge depths and a good >move ordering and a limited form of pruning, meanwhile seeing a lot with >quiescencesearch. Then programs like this play that well tactically and >search that deeply, that they have no other option to improve it other >than improving evaluation, as that is the weakest chain after you have >an equal position after book, which is another thing that'll give a lot of >trouble in the future. > >I'm seeing in the databases games of me when i was a huge rebel. Like 15 >years old and only 1900+ rated. those games are used to make my book! > >Quite laughable! > >I blundered and missed all kind of positional shots in those days! >Same for my opponents lucky, otherwise i'd never have won a game. > >Vincent
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.