Author: Robert Hyatt
Date: 06:28:03 09/28/98
Go up one level in this thread
On September 28, 1998 at 03:27:27, blass uri wrote: > >On September 27, 1998 at 20:07:10, Serge Desmarais wrote: > >>On September 21, 1998 at 19:46:15, Amir Ban wrote: >> >>>On September 21, 1998 at 18:49:25, José de Jesús García Ruvalcaba wrote: >>> >>>>On September 21, 1998 at 16:12:55, Amir Ban wrote: >>>> >>>>>On September 21, 1998 at 08:11:22, blass uri wrote: >>>>> >>>>>>Is the version for sale better than the 16bit version that I and thorsten have? >>>>>>(I am not talking about speed because it is obvious that 32bit is faster than >>>>>>16bit) >>>>>> >>>>> >>>>>You both got a version that was close to the final one but not the final one. >>>>>There were some changes added, but I don't think they affect playing strength >>>>>significantly. One change that was in the final 32-bit beta version (that played >>>>>against Zifroni) was finally not included, and the released version is almost >>>>>identical to the one that played Yudasin. >>>>> >>>>> >>>>>>Is there an option to let underpromotion(not to queen) in the version for sale? >>>>>> >>>>> >>>>>No. >>>>> >>>>>Amir >>>> >>>> I am surprised. What does it mean that there is no option in Junior5 to let >>>>underpromotion? >>> >>> >>>ok, let's be accurate: The interface allows you to underpromote when entering >>>moves, but Junior does not consider underpromotions in its search. >>> >>>Amir >> >> >>That means if the only way to draw in the endgame for Junior, it will lose only >>for not considering the ONLY legal move that would save the game? > >yes > >> And what abou >>the "typical combinations" winning a piece, like QxN+, KxQ, Pmove=N+ and >then NxQ ending up a knight? > > >Junior will not see these combinations but there are not common(The probability >that it will be relevant in a game is less than 1%) > >This "bug" was done in purpose to help Junior to save time in more than 99% of >the cases when only promotion to queen is logical. > >Uri >> >>Serge Desmarais actually it hardly saves anything at all. In 99.9% of pawn promotions, in a full-width tree search, the promoted piece is "ripped" instantly. Which means that after you search e8=Q, you *won't* have to search e8=R, because the position after Rxe8 is the same no matter what was there before, and the hash table prevents searching the nonsense. And when the promoted piece doesn't get ripped away instantly, the big material advantage of having a queen means the under-promotions get pruned away very quickly, *except* for positions where they do count. Burt Wendroff (previous post of mine) would argue with you about how frequently it comes up. It doesn't matter *how frequently* it only matters *if*. Because if you don't do it you will draw a won ending, or lose a drawn ending, or lose a won ending. I don't think it's a worthwhile optimization... All it does is make the move generation a little simpler...
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.