Author: Uri Blass
Date: 11:47:56 10/04/01
Go up one level in this thread
On October 04, 2001 at 14:23:21, Robert Hyatt wrote: >On October 04, 2001 at 00:26:38, Uri Blass wrote: > >>On October 03, 2001 at 11:30:56, Robert Hyatt wrote: >> >>>On October 03, 2001 at 09:58:48, Uri Blass wrote: >>> >>>>On October 03, 2001 at 09:47:04, Chessfun wrote: >>>> >>>>>On October 03, 2001 at 08:58:28, Uri Blass wrote: >>>>> >>>>>>I do not understand some draw decisions >>>>>> >>>>>>For example the following position from bringer's game in the last round: >>>>>> >>>>>>[D]8/2R5/8/6K1/1k6/7p/6bP/8 b - - 0 107 >> >> >> >> >> >> >> >> >> >> >> >> >>>>>> >>>>>>The last capture was at move 96 and I am not sure if white cannot win this >>>>>>position >>>>>> >>>>>>If white can get the following position in the game then it seems to me that >>>>>>black is losing(I do not have the 5 piece tablebases of KRP vs KB and someone >>>>>>with these tablebases can probably say even that it is a forced mate and exactly >>>>>>the number of moves. >>>>>> >>>>>>Uri >>>>> >>>>>It's mate in 21, but I can't see how White can force black into this position >>>>>within 18 more moves. >>>>> >>>>>Sarah. >>>> >>>>I also cannot see and the original position may be a draw with exact game >>>>If someone decide to build the 6 piece tablebases for 4 pieces with blocked >>>>pawns(white pawn at h2 and black pawn at h3) we may know. >>>> >>>>Maybe it is a good idea to tell chess programs to build these tablebases during >>>>a game when the hardware is fast enough and it is tournament time control. >>>> >>>>It may be a very good option for correspondence games even today. >>>> >>>>Even if programs need 30 to build these tablebases it may be possible to do it >>>>at some tournament time control games when this kind of position is in the root >>>>and the program may use 30 minutes for one move but later play always from the >>>>tablebases. >>>> >>>>Uri >>> >>> >>>Did you mean 30 minutes or 30 hours? 30 hours would be very fast for a 6 >>>piece file. With pawns, maybe 30 days. >> >>I did not suggest to generate the full 6 piece tablebase of KRP vs KBP but only >>the tablebases for the case when the white pawn is at h2 and the black pawn is >>at h3. >> >>the number of relevant positions to investigate is bounded by 31*61*60*59 in >>this case >> >>31 possibilities for the black bishop(it cannot be at h3 because there is a pawn >>there) >>61 possibilities for the white rook after putting the bishop >>60 possibilities for the white king after puting bishop and rook >>59 possibilities for the black king. >> >>Positions after a capture are known by the 5 piece tablebase so we do not need >>to investigate them. >> >>Uri > >That is just one case. There are _many_ others with blocked pawns. The point is that in a game you need only one of them. IE >let's make this a KQP vs KRP ending. With the pawns blocked. It turns into >a KQ vs KR for tablebase generation purposes. That takes a while... IE it >isn't computed in 1 minute or 10 minutes because there are so many iterations >to do, and the size of the temporary files are _still_ big... > >I don't think computing them on the fly is doable. It takes over 2 hours >just to compute all the 4 piece endings by themselves... Yes, but you do not need to compute all the 4 piece tablebases but only something that is sligthly harder than computing one of them(slightly harder because it is impossible to use symmetry considerations when there are pawns). if you need 30 minutes to calculate it then it may be possible in a game if you have 1 hour for the rest of the game. I also expect the time to get smaller in the future. I believe that it is also possible to build all the 6 piece tablebases with blocked pawns in the near future and it may be even a better solution(the number of possibilities for blocked pawn when you do not consider symmetry is only 20 so for every material structure building the tablebase is similiar to building a 5 piece tablebases) The number of material structures is not big and even smaller than the number of material structures for the 5 piece tablebases because 4 pieces are known(2 kings and 2 pawns and you only need to choose 2 more pieces(white and black if I assume that KPXX vs KP is usually not interesting) The number of interesting material structures when you know that 2 pawns are bloacked is exactly 15 KPP vs KPP KPN vs KPP KPB vs KPP KPR vs KPP KPQ vs KPP KPN vs KPN KPB vs KPN KPR vs KPN KPQ vs KPN KPB vs KPB KPR vs KPB KPQ vs KPB KPR vs KPR KPQ vs KPR KPQ vs KPQ 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.