Author: Uri Blass
Date: 16:07:31 09/17/00
Go up one level in this thread
On September 17, 2000 at 17:59:39, Vincent Diepeveen wrote: >On September 16, 2000 at 11:59:06, Uri Blass wrote: > >>On September 16, 2000 at 10:23:16, Vincent Diepeveen wrote: >> >>>On September 16, 2000 at 03:29:50, Uri Blass wrote: >>> >>>>On September 15, 2000 at 23:03:08, Vincent Diepeveen wrote: >>>> >>>>>On September 15, 2000 at 15:01:33, Uri Blass wrote: >>>><snipped> >>>>>>In order to solve chess you do not need tablebases of all the legal position but >>>>>>only tablebases of all the logical position. >>>>> >>>>>>You do not need to analyze illogical lines like 1.e4 f6 2.e5 g5 3.e6 in order to >>>>>>solve chess because it is clear that you never reach this position in practical >>>>>>game. >>>>> >>>>>This is bullshit. >>>>> >>>>>A program isn't smart enough to know e4 f6 is nonsense. >>>> >>>>It can be smart enough to not analyze 1.e4 f6 2.e5 g5 3.e6 because 3.Qh5# is >>>>better. >>>> >>>>I did not say that it is smart enough to know that 1...f6 is illogical and if >>>>1.e4 is the first move it needs to know how to play after 1.e4 f6 so it does >>>>need to analyze 1.e4 f6 but it does not need to analyze the line that I gave >>>>because 2...g5 is illogical move and 3.e6 is another illogical move and there is >>>>no point in knowing the right move in cases that both sides did mistakes because >>>>these positions will never happen in games. >>>> >>>>Uri >>> >>>You still have problems understanding how search works in computerchess. >>> >>>Obviously e6 doesn't get analyzed if Qh5 gives a cutoff already, >>>but all other nonsense moves in our eyes like e4 f6 d4 a5 and such they >>>sure need to get analyzed further, so many nonsense lines in our eyes >>>need to get analyzed till the far end. >> >>There is a misunderstanding. >>I understand how search work in computer chess. >>The discussion was about the complexity of the chess game. >> >>I suggested one definition as the size of the tablebases that you need in order >>to solve the game(not by search but by tablebases). >> >>I know that programs use the alpha beta but in order to build tablebases you do >>not use the alpha beta. >> >>I said that you do not have to have an evaluation for all the legal positions in >>order to do it but only for part of them and it is possible that you need bigger >>tables in order to solve another legal position. >> >>I know that the part is so big that it is not practical but if you have enough >>memory and enough speed you can start by finding all the pseudo logical games of >>1 ply,2 plies,3plies.... and building the tablebases(starting by finding all the >>mate positions). >> >>The program that build the tablebases in this case can see that 1.e4 f6 2.e5 g5 >>3.e6 is illogical because both sides did mistakes so it can ignore this position >>and positions after it. >> >>Uri > >EGTB don't work like that. You store all legal positions, including mate in 0 >positions (so positions where you are mated). You need this in order to >to see that other positions are mate in 1, 2, .... n I agree that EGTB today do not work in this way but if we want to solve chess we can save space by building them in a different way. First we need to store all the logical positions(a position that we get when one side missed a forced mate is clearly illogical) after 1 ply,2 plies,3 plies,... when we stop when we do not get new positions. After it we can find all the mate in 0,mate in 1... > >EGTB don't work with lines of moves, so if it gets in the position >after e4 f6 e5 g5 e6, then it won't get helped. > >It's just busy constructing mates in n+1 after having passed mate in n. > >So it's busy doing 1 move exactly in a position and doesn't store the moves >as that would eat another 12 bits or so. Waste of memory and useless to >store! > >Note that this all is theoretical bullshit anyway, as chessprograms >will perfect chess long before any EGTB will, as they don't need to >search 10^40 or something positions to solve it, but way less! I agree that chess programs do not need to search all the positions to solve chess. They even need less than tablebases because they do not need to know exactly what to do in every case(if they find that 1.e4 is leading to a draw and also find that 1.d4 Nf6 is leading to at least a draw for black they do not need to know if 1.d4 Nf6 is a draw or a win for black and it means that they do not need to know the best move against 1.d4). tablebases that tell the distance to mate in every practical position need more positions even if they do not tell the distance to mate for every legal position. 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.