Author: Peter Fendrich
Date: 16:20:10 08/14/03
Go up one level in this thread
On August 14, 2003 at 18:32:01, Uri Blass wrote: I'm sorry if I mislead you with the sloppy defintion of 1FH. See my answear to Dieter. Do you still have the same opinion? I think it's a good measurement but it has to be backed up by others of course. The first move is of special interest I think. The order of the second, third etc is of less importance but still interesting. The order of (let's say) move 10 and up is not at all that interesting for the overall performance. /Peter >On August 14, 2003 at 18:02:16, Peter Fendrich wrote: > >>On August 14, 2003 at 16:18:50, Gian-Carlo Pascutto wrote: >> >>>On August 14, 2003 at 15:27:06, Peter Fendrich wrote: >>> >>>>1FH = The ratio where the first generated move in the node is a Fail High. >>>> >>>>I experimented with removing the the hash table and became a bit surprised. >>>>Normally I have about 1FH = 95-96%. When I removed the hash table that figure >>>>raised to more like 1FH = 96-97% while the search depth, as expected, was >>>>decreased. >>>> >>>>Increased 1FH when removing the hash table. Is this normal? >>> >>>In my program, move ordering worses when I disable hashing. >> >>Well, it should be the same in mine program as well and the reduced search depth >>implies that this really is the case. >> >>>Do you handle the case where there is no reliable move in the hash? >> >>This is why I started to do the test in the first place. >>I wanted to test a new idea instead of IID which is quite expensive. For this >>test I disabled both the new idea and IID. Left was only good captures, Killer >>moves and other well known methods to sort the moves. >>Without the hash moves the search tree becomes bigger allright. Maybe it's >>possible to get higher values for 1FH and still have a bigger search tree and >>worse move ordering. I don't understand how, though... >> >>/Peter >> > >It is clearly possible to get higher values for 1FH and bigger search tree. > >the test if first fail high is a bad test and I never use it. > >The test tells you the following wrong information: > >1)If your first move did not fail high than it is better to have bad order of >the rest of the moves because if you have bad order of the rest of the moves you >have good chances that the first move is going to fail high against them. > >If you have good order of moves than the you get no points for the fact that >your second move failed high. > >2)It is not important if you start with a mate or with a move that win a pawn >inspite of the fact that mate means tree of 1 node when starting with first move >that win a pawn may lead to a big tree. > >It is even possible that starting with winning a pawn is going to be considered >as better because you may get first move fail high later in the search. > >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.