Author: Robert Hyatt
Date: 14:59:27 06/22/00
Go up one level in this thread
On June 22, 2000 at 16:19:18, Mogens Larsen wrote: >On June 22, 2000 at 13:59:15, Robert Hyatt wrote: > >>Nope. even on wrong predictions there are _many_ transpositions that make the >>incorrect pondering entries work. IE I ponder expecting you to play Nf3, but >>you play e3 instead. However, when I expected you to play Nf3, I discovered >>that your next move would be e3. Searching NF3/e3 in that order will then >>result in transpositions when you try e3 first, as e3/Nf3 is still the same >>thing. > >Yes, but you would only discover the move e3 after it's made and then you're >using time in transistion from Nf3 to e3 or am I missing something? Unless you >happened to ponder e3 intensively before switching to Nf3, but since you >overwrite old data then... > Remember the tree. If I play Nf3, you play e6, I play Ng5, I reach the same position as if I play Na3, you play e6, I play Ng5. Totally different first move, totally identical final position. The hit (hash) after Ng5 works no matter which move is searched first. >>Also I know the difference between old positions and new positions, and I always >>choose to overwrite positions from an old search before I overwrite something >>from the current search. > >Wouldn't squeezing new data in before overwriting old data be better just in >case it's usable. Or could that cause problems? It does that. But at reasonable time controls, the hash table gets overwritten many times during a single search. Remember that I search about 1M nodes per second. that is 300M in 5 minutes. at 16 bytes per entry, that would require 4.8 gigs of RAM. Very few have that much RAM for hashing. :) If you cut that by a factor of 10, then the tree is 10x bigger than hash. Overwrite city. > >Best wishes... >Mogens
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.