Author: Andrew Williams
Date: 11:10:17 06/10/02
Go up one level in this thread
On June 10, 2002 at 13:44:26, Les Fernandez wrote: >On June 10, 2002 at 13:15:55, Andrew Williams wrote: > >>On June 10, 2002 at 11:18:25, Les Fernandez wrote: >> >>>On June 10, 2002 at 11:01:50, Robert Hyatt wrote: >>> >>>>On June 10, 2002 at 03:22:55, Andrew Williams wrote: >>>> >>>>>On June 10, 2002 at 00:18:51, Les Fernandez wrote: >>>>> >>>>>>On June 10, 2002 at 00:02:39, Robert Hyatt wrote: >>>>>> >>>>>>>On June 09, 2002 at 23:45:38, Les Fernandez wrote: >>>>>>> >>>>>>>>I rarely come across strange output using Crafty but this position doesnt seem >>>>>>>>to work right. The ce after analysis I think should be -32765 but in fact it is >>>>>>>>reporting a "0"??? Either I am dead tired or something is screwy. Would >>>>>>>>somebody using Crafty run this epd string? It is a very simple mate and I cant >>>>>>>>seem to find the problem. Its probably just late <S>. >>>>>>>> >>>>>>>>Here is Crafty analysis: >>>>>>>> >>>>>>>>rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - acd 1; acn 1; acs 1; ce 0; >>>>>>>>pv Rh7; >>>>>>>> >>>>>>>>The pv appears to be right but the ce value is dead wrong!! >>>>>>>> >>>>>>>>If you find the same results can someone please explain the nature of this? >>>>>>>> >>>>>>>>[D]rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - >>>>>>>> >>>>>>>> >>>>>>>>Thanks >>>>>>> >>>>>>>Can you describe _exactly_ how you produced that? IE post the epd file >>>>>>>you used, etc... >>>>>>> >>>>>> >>>>>>EPD Kit revision date: 1996.04.21 >>>>>>unable to open book file [./book.bin]. >>>>>>book is disabled >>>>>>unable to open book file [./books.bin]. >>>>>>search time set to 180.00. >>>>>>hash table memory = 48M bytes. >>>>>>pawn hash table memory = 10M bytes. >>>>>>Crafty 18.9 vs ferret >>>>>>kibitz Hello from Crafty v18.9! >>>>>>pondering enabled. >>>>>>play best book line after search. >>>>>>EGTB access enabled >>>>>>using tbpath=./TB >>>>>>4 piece tablebase files found >>>>>>1406kb of RAM used for TB indices and decompression tables >>>>>>PFGA: EPD record: 1 >>>>>> >>>>>>The exact epd string that was in epd.epd was: >>>>>> >>>>>>rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - >>>>>> >>>>>>And the epd.out file was: >>>>>> >>>>>>rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - acd 1; acn 1; acs 1; ce 0; >>>>>>pv Rh7; >>>>>> >>>>>> >>>>>>The analysis report can be found in my original post here. This one puzzles me >>>>>>too Bob but I could not rule out that its late <S>. If you need any additional >>>>>>info let me know. >>>>>> >>>>>>Thanks >>>>> >>>>>Is this because Black has only one move? In my program, if there's only one move >>>>>available *during a game* it only searches to depth 4. For testing epd >>>>>positions, it searches normally, even if there is only one move. >>>>> >>>>>Andrew >>>> >>>> >>>>That might be the reason. With only one move, the epd code in Crafty might >>>>not do a search at all, and that would leave the score unchanged from the last >>>>epd position perhaps... >>> >>>Um I am not sure that is what is going on here, let me explain. What I did was >>>I took an epd position that reported a mate sequence and then made the pv moves >>>and then created all the following epd positions that eventually led to the >>>reported mate based on the pv sequence from the first position. Following is >>>the results of Crafty which ran the epd's that I created based on the pv line >>>from the first position: >>> >>>rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/R7 w - - acd 4; acn 10909; acs 1; ce >>>32764; pv Rh1+ Rh7 Rxh7#; >>>rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - acd 1; acn 10909; acs 1; ce >>>0; pv Rh7; >>>rnb4k/7r/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R w - - acd 4; acn 7328; acs 1; ce >>>32766; pv Rxh7#; >>>rnb4k/7R/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/8 b - - acd 1; acn 1; acs 1; ce >>>-32767; pv; >>> >>>As you can see the second epd string reports a "0" but the one before it and >>>after it are in fact reporting a value. If your thought "With only one move, >>>the epd ..... " is correct then why would the last epd position which is the >>>actual final move show a ce of -32767 ??? Any other thoughts Bob? >>> >>>Thanks >> >>I think the explanation is correct. The only position you listed which has >>*exactly* one legal move is the second one, the one where crafty reports 0. The >>problem is probably that as a way of saving time in games, many programs just >>make the move without searching it (because there's no choice). For the last >>position, there are *zero* legal moves. Presumably crafty treats this as a >>special case as well (my engine does). If there are no legal moves and the side >>to move is in check my engine returns -M00. If it's not in check, it returns >>DRAWSCORE. I can't be sure if this applies to crafty too, but I think it's >>consistent with the original explanation. > >Hi Andrew, > >I see what you are saying but doesn't this pose a problem. ie suppose I have an >engine which is doing some type of lookup and sees the ce=0 versuses a mate >announcement. If what you are saying is correct does that mean we will never >see a ce value = -32765 ??? In the past I know I have gone through millions of >epd strings from various sources with the intention of extracting only those epd >strings that are end in mate. I look up ce values and only keep those positions >that have an abs(ce)>=32000. How do I get around this delemna? I think it depends on the engine that created the epd strings. For example, mine produces this for the problem position above: rnb4k/5r2/2p1p1Q1/pp3pP1/3P4/b2B1KP1/1PPB4/7R b - - acd 13; acn 1657088; acs 10; ce -32765; fmvn 1; hmvc 0; id "fernandez.epd.0001"; pm Rh7 ; pv 1...Rh7 2.Qxh7 ; But I guess it's up to the individual programmer how to handle these cases. If you're not proficient in C yourself, perhaps you might ask if anyone else wants to make necessary change to crafty for you? Andrew PS I've got about 4500 of these from PM's games if you want them. I should warn you that they may not all be correct.
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.