Author: Robert Hyatt
Date: 18:22:55 09/26/01
Go up one level in this thread
On September 26, 2001 at 15:32:38, Gian-Carlo Pascutto wrote: >On September 26, 2001 at 14:56:02, Robert Hyatt wrote: > >>The last thing is interesting. Still has my curiousity up to see just how >>many "errors" are required before the search falls apart. > >I agree it would be quite interesting. > >It's not actually so surprising. After all, _all_ our evaluations are >'incorrect' to some extent. We use the tree search to help with this. >But the tree search itself can also be pathological. > >Generally, the closer an evaluation is to the leaves the more 'error' >we can sustain. And because of the exponential nature of the tree most >of them will be just there! > >Before getting a serious error at the root you would have to get >a serious number of leaf or near-leaf nodes to 'conspire' (the analogy >should be obvious...) and give bogus scores. > >Even if the search falls apart it would still have to do so in >a way that actually produces a bad move at the root. Having a bogus >score is annoying but won't kill you in games. Putting your queen >en prise will. > >I don't mind generating several collisions per second. Why do so >when I generate 100 000 'erronous' evals per second? > >-- >GCP I won't argue about erroneous evals. But at least they should be consistent. Here we are talking about something that will be highly inconsistent, where we "sniff" a fail high but then can't prove it... etc...
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.