Author: martin fierz
Date: 13:33:40 09/28/04
Go up one level in this thread
On September 28, 2004 at 14:59:37, Robert Hyatt wrote: >On September 28, 2004 at 13:59:58, Steve Maughan wrote: > >>Bob, >> >>>What I do is this: When I get a hash hit, and the draft is not enough to let >>>me stop the search at that point, I then test the table draft against the >>>depth I would use for a null move search at this point. If the table draft is >= that depth, and the table entry says "No way I would fail high here" then >>>there is no point in trying a null-move search, if a normal move search would >>>not fail high... >> >>Interesting - I know that you alluded to this in the past but this is the first >>time the penny has dropped for me as to exactly what you're doing. >> >>So to use example numbers: >> >>Alpha = 100 >>Beta = 101 >>Depth to Go = 5 ply >> >>Hash Score = 50 >>Hash Bound = UPPER >>Hash Draft = 4 ply >> >>In this case you'd skip the null move - seems sensible. >> >>Do you know how much is it worth in speedup? >> >>Thanks for the explanation, >> >>Steve > > >Since WAC141 has been a hot topic, here is crafty with the normal avoid-null >stuff as I explained: > > 9 3.95 -1.20 1. Kf1 a5 2. Rxh5 gxh5 3. Rxh5 Re1+ > 4. Kxe1 Nxh5 5. Qg5+ Kf8 6. Qh6+ Kg8 > 7. Qxh5 > 9 4.42 +1 1. Qxf4!! > 9 4.67 +3 1. Qxf4!! > 9 5.16 +M 1. Qxf4!! > 9 50.78 Mat06 1. Qxf4 Bxf4 2. Rxh5 gxh5 3. Rxh5 Bh6 > 4. Rxh6 Qh2+ 5. Kxh2 Kf8 6. Rh8# > 9-> 50.78 Mat06 1. Qxf4 Bxf4 2. Rxh5 gxh5 3. Rxh5 Bh6 > 4. Rxh6 Qh2+ 5. Kxh2 Kf8 6. Rh8# > time=50.78 cpu=99% mat=-1 n=79749173 fh=98% nps=1.57M > > >Here is same thing but without the avoid-null stuff so it always tries a null >move: > > 9 3.93 -1.20 1. Kf1 a5 2. Rxh5 gxh5 3. Rxh5 Re1+ > 4. Kxe1 Nxh5 5. Qg5+ Kf8 6. Qh6+ Kg8 > 7. Qxh5 > 9 4.40 +1 1. Qxf4!! > 9 4.75 +3 1. Qxf4!! > 9 5.22 +M 1. Qxf4!! > 9 49.51 Mat06 1. Qxf4 Bxf4 2. Rxh5 gxh5 3. Rxh5 Bh6 > 4. Rxh6 Qh2+ 5. Kxh2 Re2 6. Rh8# > 9-> 49.51 Mat06 1. Qxf4 Bxf4 2. Rxh5 gxh5 3. Rxh5 Bh6 > 4. Rxh6 Qh2+ 5. Kxh2 Re2 6. Rh8# > time=49.51 cpu=99% mat=-1 n=80501452 fh=98% nps=1.63M > >So about 1.5% better here. Other positions might be even better... > >Note ignore the time and just look at total nodes. second run was not a full >PGO compile... just a very short correction: it's 1.0% better, not 1.5%, do you still have one of those P4's with math bugs? :-) cheers martin
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.