Author: Dave Gomboc
Date: 21:17:27 08/10/99
Go up one level in this thread
On August 10, 1999 at 21:55:29, Robert Hyatt wrote: >On August 10, 1999 at 18:42:58, Tom King wrote: > >>On August 09, 1999 at 20:41:50, Robert Hyatt wrote: >> >>>On August 09, 1999 at 17:51:01, Tom King wrote: >>> >>>>On August 08, 1999 at 21:35:46, Robert Hyatt wrote: >>>> >>>>[snip] >>>>> >>>>> >>>>>I've been doing a 'hybrid' for nearly a year... R=3 fairly close to the >>>>>root, R=2 beyond some threshold that I don't remember what it is... >>>>> >>>>> >>>> >>>>Interesting.. better than pure R=2 or pure R=3? I did some experiments a while >>>>back, using R=3 on the opponent's move, and R=2 on the computers move (idea >>>>being - let's not cover up the opponent's threats, and if we miss some of our >>>>threats, well, so be it). But it was more or less a 'wash'. >>> >>> >>>I started playing with this after Paris, as someone said that Frans (or someone) >>>was fiddling around with R=3. Bruce and I played a few games one night with >>>Ferret vs Crafty, with bruce using R=3, and we saw _no_ gross mistakes on his >>>part. In fact, we couldn't tell the difference (at least R=3 was not making >>>mistakes that R=2 wasn't). >>> >>>But I never felt comfortable with raw R=3, as R=2 causes more than enough >>>problems already, thank you very much. :) >>> >>>Ernst is writting an ICCA paper describing some experiments he did with this >>>(totally unconnected with my tests). Wait for his paper to hit the journal as >>>he has some good data... And I am not yet sure that this is a good idea, but I >>>have been running it a good while and it has been in the distributed Crafty for >>>quite a while as well and no one has complained. Of course, now that everyone >>>knows, look out. :) >> >>One thing's for sure - I'm sure there's a lot more we could all be doing with >>the null move. Playing around with different values of R, depending on depth in >>the tree, alpha and beta etc. etc. I look forward to this paper. Ernst has been >>productive of late..I'm sure I'm not the only one to enjoy reading his articles. >>[even if his damn program does kill mine at the WCCC ;-)] >> >>Cheers, >>Tom. > > >If you want to play around, here's another idea I have on my 'to-do' list: > >at present I 'break' the search into two 'chunks'. the part near the root >uses R=3, the rest uses R=2. Something tells me this might be made much more >dynamic than that... ie R=4, then 3, and finally 2. But rather than some >static divisor as I have now, make this dynamic so as you go deeper, you use >bigger R values near the root, etc... > >Seems reasonable. Whether it will work or not, we won't know until we try >it... What relationship, if any, would you posit between a good R and how sharp or quiet a position is? Dave
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.