Author: José Carlos
Date: 06:57:19 05/27/04
Go up one level in this thread
On May 27, 2004 at 07:42:28, Sune Fischer wrote: >On May 27, 2004 at 04:25:10, José Carlos wrote: > >>On May 26, 2004 at 15:53:45, Sune Fischer wrote: >> >>>On May 26, 2004 at 14:43:14, Robert Hyatt wrote: >>>> >>>>It isn't so easy for me. Crafty ponders automatically unless it is turned off. >>>>If it is turned off, it never ponders. It decides which move to ponder by >>>>itself, and has code to do that pretty well. UCI wants to handle everything. >>>>IE the engine is a small part of the whole "thing" while Crafty considers the >>>>engine to be _the_ whole thing. IE it handles its own book, book learning, >>>>needs to know when/how a game ends, etc. Crafty correctly claims draws, mates, >>>>resigns, and all of that. >>>> >>>>To disable all of that to work with UCI is simply not worth the effort, when the >>>>winboard protocol works just fine and has for years. >>>> >>>>IE at the root, I want to decide whether a position is a tablebase draw or not, >>>>and use my "swindle mode" if it is. I can't do that in UCI. There are too many >>>>things I can't do, or which I have to drastically change, to make it work. >>>> >>>>"If it ain't broke, don't fix it" is my motto here. :) >>> >>>I am of the same basic ideology, but in all fairness UCI does have a few sweet >>>features. >>> >>>My favorite ones: >>> >>>1) the engine can show the current line of thinking. Pretty easy to implement >>>too, just print the moves leading from the root to the current node. >>> >>>2) multi pv, display scores and pv for the n-best moves >>> >>>3) refutation lines, show how the opponent refuted the root move dxe5!?. >>>I like this one a lot, finally the user can become "educated" by the engine as >>>it now shows you why things _don't_ work! :) >> >> I don't know how UCI helps in this. To be true, I haven't read the UCI spec >>since it was first presented. >> But I show this refutation in winboard with no problem. When I fail low at the >>root, I read the whole line from the hash table (I save a move even in fail >>lows, not for move ordering, but to have it always ready to print). I did this >>after my latest release of Averno, so it's not in the public version. >> Maybe I'm misunderstanding what this "refutation" means... >> >> José C. > >I don't think that's a real refutation line, if you fail-low it just means the >score for the PV move is lower than your initial aspiration window (right?). > >If you later fail-high on another root move, then you haven't really "refuted" >the first move, but merely found a better move. That's how I'd use the word >anyway. Refuted means the opponent failed-high on your move. I didn't mean this. I meant: when I fail low at the pv move at the root, I print the complete variation, being the second move the "refuting move" for my pv, and then the rest of the line shows the final position where I found I was worse than I thought. For example, I plan to castle kingside and then find Bxh7 winning for the opponent. Then my program prints something like 1...0-0 2. Bxh7+ Kxh7 3. Ng5+ Kg8 4. Qh5 etc. so I can see why I failed low. >The refutation lines, as I do them, is about printing the line for every move >that didn't make it to a PV. > >E.g. let's say you have position with 20 root moves, the first one searched is >the PV move and you find a score for this move whatever the score may be, then >you go on to the next move and this fails-low (as expected), but _why_ did it >fail-low? > >That's where I print the refuation line, ie. the opponent must have had a move >that made him fail-high on this move, what was this move? > >So, assuming your first root move holds you will have 1 PV line and 19 refuation >lines. Of course these lines will be fulll of nullmoves and other strange >things, but at least one can see what the engine is "thinking" about the other >moves. > >I believe that's what's meant in the UCI protocol, though I'm not entirely sure >because it's not explained in detail. > >-S. I understand. That's very interesting and similar to what old Fritz (I don't know if newer Fritz still does this) did showing pv's for the n best moves. Many times those pv's where simply refutation lines (not really pv's with exact score). Thanks for the exaplanation. José C.
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.