Author: Vincent Diepeveen
Date: 14:15:28 12/08/99
Go up one level in this thread
On December 08, 1999 at 15:28:46, Eugene Nalimov wrote:
>Main benefit from hashing is not cutoff of the entire line - that happens rare
>in the middlegame (let's say speedup will be ~10%). Main benefit is that you
>often get the best move from the shallow search, and there are good chances that
>that move will be good for deeper search. Even slightly better moves ordering
>helps alphabeta a lot.
>
>And of course it's better to have iterative deepening as well.
>Eugene
For the average program getting only a few ply at 486,
this might be true, but i'm sure the top
programs at depths they search nowadays this is no longer
true.
Main benefit for DIEP from hashtable is
- 60% of all evaluations out of hash, only 40% need a full eval.
- transposition table, nowadays programs search a lot deeper than
in the past
My move ordering because of hashtable hardly get better (so the
chance that the first move gives a cutoff or improves alfa or beta),
just a few tenth of % up to 1% at maximum.
This is because of the nature of nullmove and the beta
pruning in the quiescencesearch, which returns usual only bounds.
Especially in my draughtsprogram main win from hashtable is
the transposition (no eval table there hits % is too low to use one),
that really kicks butt there.
>On December 08, 1999 at 15:10:16, leonid wrote:
>>On December 08, 1999 at 13:25:15, Vincent Diepeveen wrote:
>>>On December 08, 1999 at 10:45:15, leonid wrote:
>>>>On December 08, 1999 at 10:17:47, Robert Hyatt wrote:
>>>>
>>>>>On December 08, 1999 at 08:32:48, leonid wrote:
>>>>>
>>>>>>Hello!
>>>>>>
>>>>>>How many plies chess game should have before hash tables could be of some use in
>>>>>>the chess game?
>>>>>>
>>>>>>Thanks,
>>>>>>Leonid.
>>>>>
>>>>>
>>>>>Three + captures should do it:
>>>>>
>>>>>Nh3 xxx Ng5
>>>>>Nf3 xxx Ng5
>>>>>
>>>>>and you avoid the rest of the 2nd search...
>>>>
>>>>Thanks!
>>>>
>>>>I tryed up to 5 plys and had no improvement. Something wrong with my logic. Will
>>>>go back and try 8 plies deep.
>>>
>>>there are always bugs in hashings... ...test it well, create at least
>>>a functoin that calculates the hashkey in the current position, and
>>>compare that with the incremental value. I bet there are some mistakes
>>>there!
>>>
>>>After that check go figure out how you store it. I've had nasty bugs
>>>when compressing a lot of info into a few bits.
>>
>>
>>I see that what you said have a good sense. I already killed one bug and rewrote
>>my hash logic for plies from 7 to 5. Still no speeding. Probably some bug more.
>>It sound so strange, since this part of logic apparently is so simple. Will
>>later go down to the minimax and see the difference. Alpha-beta can obscure
>>efficency of hashing.
>>
>>Leonid.
>>
>>
>>>
>>>>Leonid.
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.