Author: Vincent Diepeveen
Date: 14:19:50 10/13/02
Go up one level in this thread
On October 13, 2002 at 16:55:32, Jeremiah Penery wrote: first of all. in endgames when i search dual i get huge mainlines. sometimes 25 ply and i have a nominal search depth of like 15 then. if i add a bunch of processors i sometimes get weird long mainlines because it searches nonsense. this is a well known phenomena. now they search very small search depths with a big number of processors. So they all search incredible nonsense. also we know they extended a lot for not so common reasons. that explains obviously a lot. so the only reliable mainlines is the first mainlines of a game. we just have 6 games as you know. after that the hashtable is filled with nonsense of a few ply which adds up quickly. extensions do the rest of the job. i do not understand how just the statement of 1 person could cause so much confusion. >On October 13, 2002 at 15:55:05, Vincent Diepeveen wrote: > >>On October 13, 2002 at 15:47:43, Jeremiah Penery wrote: >> >>>On October 13, 2002 at 14:14:13, Vincent Diepeveen wrote: >>> >>>>On October 13, 2002 at 13:50:24, Jeremiah Penery wrote: >>>> >>>>for any chessprogrammer it is very clear Jeremiah. >>>> >>>>It's 12 ply and the 6 only means that the hardware search can >>>>be up to 6 ply within that 12 ply nominal. >>> >>>For the millionth time, explain 4(5) then. >> >>it's a lot easier when you try to imagine how you would start >>a search. Look at 4 ply you want to involve other cpu's too. >> >>You have 480 chess chips. 9 out of 10 is idling anyway >>(we know this because everyone says it was capable in theory >>to get a billion nodes a second in theory, yet it got only >>126 million nodes on average; there is no discussion here) >> >>you want to use as many cpu's as possible. So at the root you >>ask a few chips. Of course you must take into account extensions too. > >If they needed an exact score from the chess processors, they could do the same >search on many processors at once with different bounds, because the hardware >searched with a null-window. So it's like MTD researching to converge on the >real value, with the potential to take only the amount of time of a single >research, also having the side effect of using lots of processors, if that's >what your goal is. > >>If a 4 ply search for this chip is too much, then you split it. and >>split it. Giving a 2 ply search to chip A, and a 3 ply search to >>chip B. > >If you want a 4-ply search, how do you get it from a 2-ply plus a 3-ply search >on different processors? > >>The 5 in short has no meaning here. > >No, it means the hardware search (not counting quiescense or extensions) could >be up to 5 ply. :) What it means in relation to the first number, however, is >not so clear. >But you said above about 12(6), "It's 12 ply and the 6 only means that the >hardware search can be up to 6 ply ***within that 12 ply nominal.***" If 5 has >no meaning in 4(5), then your previous statement was simply wrong. Else, the >meaning of 4 is still not completely clear. > >>You can see that the mainlines at the start of the game (so before >>hashtables are filled with a lot of info that extends lines) >>that the lines it gives at 4 , 5 , 6 ply is usually 1 ply long. > >There is hardly ever a 1-ply line given. After a wrong pondering guess, I often >see depth of 3(4), and the mainline is almost always 3-5 ply given, sometimes >much longer. Sometimes after a wrong pondering, the first depth given is more >like 8(4), with line of about 8 ply. 5(5) is also a common number, and >sometimes the lines are 1 ply, but more often 3 or more. > >If you want, I can go through all the logs later today and make a table of the >first search output after a wrong pondering, with the depth - x(x) - and the >length of the given PV.
This page took 0.01 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.