Author: José de Jesús García Ruvalcaba
Date: 15:01:16 06/22/99
Go up one level in this thread
On June 22, 1999 at 09:50:26, Vincent Diepeveen wrote: >On June 21, 1999 at 14:34:20, José de Jesús García Ruvalcaba wrote: > >>On June 21, 1999 at 13:53:48, James Robertson wrote: >> >>>My program does check, recapture, and pawn push extensions, but I have heard >>>that many programs do a lot more. I think Gromit said he does something like 10 >>>different extensions? >>> >>>Could someone please tell me what different kinds of extensions are sometimes >>>used? It would be cool to try them out on my program. >>> >>>James >> >>1. Pawn ending extension. >Extension invented by Walter Ravenek. >i'm already searching like 40 ply in any pawn ending, >but the thought is obvious; if you exchange TO a pawn ending >then every move gets extended. you see quickly when you can >exchange to a pawn ending then. a lot of programmers are nowadays >doing it. > >>2. Fail high extension. >how can you call that an extension! > >>3. Preferred variation extension. >that's crap. >>4. One-reply extension. >>5. Singular extension. >>6. Enabled move extension. >enabled move extension that asks for more explanation! > This is used when a move is enable by the previous move by the same player. For example, if in the initial position white plays 1. e4, then for white's next turn the queen's and light-squared bishop moves are extended. >> I have an idea; but I have not tried it as I am not a chess programmer. Extend >>on moves which drastically (this has to be defined) increase the mobility and/or >>decrease opponent's mobility. Moves which do the opposite can be forward-pruned, >>too. > >What is written above is crap. > Let me guess: you have already tried it and it did not work (: That is not surprising. But it is interesting that this "mobility extension" has big overlaps with check extensions, recapture extensions and one-reply extensions. >To describe most common extensions; > > - check extension > always, or limit it. Why extend every check like e4,e5 Qh5,Nc6 Qxf7?? > - threat extension; if nullmove returns threat then extend. don't extend > all times nullmove returns threat. most common is to extend when > nullmove returns mate score. I call that mate threat extension. I limit > this a lot, yet this extension still works great. > >- recapture extensions; this never brought me anything, yet most programs > are doing it. i guess it's depending upon what you try in qsearch. the > more accurate your qsearch is the less you need of these is my thought. > i don't do them > >- singular extensions; i've tried this thousands of times, but never found > them working. A singular extension is in fact a case of threat extension; > there is a threat that only allows a single move to prevent it by a > margin S. Threat extension doesn't have problems with S, and also extends > when more than 1 move prevents the threat. > >- passed pawn push extensions; i'm doing a few of those. can blow up your > search awful sometimes i limit it considerable. > >- one reply extension; when you only have 1 legal move, then extend it >independant from other extensions you already have in that pos (like check) > >- check evasion; after giving a lot of checks, it might be worth giving some > more checks (as a kind of selective search to find a checking sequence that > might mate). > >- high score extension. To prevent horizon effects to occur, one can > extend if a certain pattern is on the board which gives a high score > in evaluation. For example: if a queen is attacked, and you give a high > score for a queen being attacked, then you might want to extend it, > resulting in a better leaf score. Of course this extension needs to be > done near leafs and not in the middle of the search.
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.