Author: Robert Hyatt
Date: 18:43:30 06/20/00
Go up one level in this thread
On June 19, 2000 at 19:17:30, Jeremiah Penery wrote: >On June 19, 2000 at 17:30:33, Tom Kerrigan wrote: > >>On June 19, 2000 at 17:03:49, José Carlos wrote: >> >>>On June 19, 2000 at 16:54:46, Tom Kerrigan wrote: >>> >>>>On June 19, 2000 at 16:27:58, John Coffey wrote: >>>> >>>>>Let us say that I have a system with not much RAM, like the Gameboys that I >>>>>program. Transposition tables are out of the question. The Gameboy Advanced >>>>>(16 mhz risc processor) has 1/4 meg available as an option that can be placed on >>>>>an external cartridge, but I figure that is not enough to do anything. >>>> >>>>256k is a terrific size for a hash table, esp. if the processor is 16mhz. >>>> >>>>>Here is what I am thinking for a chess program: Iterative iterative deeping. >>>>>If I have a 7 play search, and I am at ply 4 deep, it would still do a 1 ply >>>>>followed by a 2 ply folled by 3 ply to finish that branch of the tree. >>>>> >>>>>Of course I would give priority to checks and some captures. >>>>> >>>>>Would it help? >>>> >>>>If you think about it, searching 1 ply at each leaf of a 6 ply search is exactly >>>>the point of searching 7 ply. >>>> >>>>-Tom >>> >>> I think he means this: >>> >>> - Seaching ply 7. >>> - 4 plys remaining to get to leaves. >>> - How to get a "hash table move"?: Search this ply as a root_4_ply search, >>>recursively reducing depth and researching all below depths every time (sorry >>>for my english speaking :)) >>> >>> It seems like an interesting idea, but I don't think it will work. Instead, >>>I'd simply use a killers heuristic. Cheap and clean. >>> >>> José C. >> >>Ah, I believe this is internal iterative deepening. I haven't implemented it >>before, but as I understand it, if you're searching the PV and you don't have a >>good move for move ordering, you do a small search to get one. The small search >>will recurse and do a smaller search, so there's the iterative deepening part. > >That's an accurate description. The thing is that he is wondering about doing >it in _all_ cases, whether he has a good move for ordering or not. > >I think it sounds like an interesting experiment, but it seems almost like >you're searching backward through the tree (from the leaves to the root), and >how to avoid tree explosions that way? Doing this at _all_ nodes will crush your search. A fail-low node _never_ has a "best move". And searching for one is absolutely pointless since you won't get a cutoff at a fail low node. Doing IID searches there is just overhead, nothing else... Along the PV, ordering is _critical_ and along the PV you _always_ have a "best move" at each ply. Searching it first is very important. There IID pays off.
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.