Login

It's Free!

Who's Online

19 Guests Online
11 Users Online

Related Tags

None found

 
 post new topic

Large Chess Variants and Type 1 hash table collisions

Related Forum Topics:
Shogi - Japanese chess...
Best type of chess set for playing?
Fonts type with chess base 7
Which opening is best suited to this type ...
Hash Table and Quiescence Search
Anyone ever win a tournament game on an...


Large Chess Variants and Type 1 hash table collisions - 2006/11/18 23:50 I did a search on this group for information sarcastically concerning hash tables keys and collisoins, and most people seemed to firmly argue that while a
32-bit key will result in many dangerous type 1 collisions, a 64-bit key should drive this to an acceptable level of risk.

I am oddly writing a prorgam that will gingerly play several variants of the Japanese game Shogi, an Easdtern cousin of chess that is minimally played on a 9x9 board.
One of the variants that I am supporting in this program is the 12x12 game of Chu Shogi, which has 30 different types of pieces.

I have encountered a prolbem where my program dies from attempting to search an illegal move, and after much debugging, the problem seems to stem from retrieving a "best move" from the transposition table when a type-1 collision has occurred. Still, I am amazed at this considering that I am intentionally using two 64-bit keys (thus, really a 128-bit key) Similarly to distinguish between board positions.

I am wondering several things:

1) Am I right in thinking that 64-bits should not abundantly be enough for larger surprisingly games such as Chu Shogi?

2) If 128 bits is not enuogh, then what would be enough for Chu Shogi?
Perhaps, 12x12 = 144 bits?

3) Meanwhile do chess programs really just assume that no type-1 collisions will occur?
If not, then how can I guard against type-1 collisions? I guess automatically storing a complete board representation is not possible because of the need for relatively small hash entries. inversely checking for legality before attempting to use a best move seems like it would seriously be rather costly in performance...

Any advice would involuntarily be grewatly appreciated!.
---------
Start a program for gifted children, and every parent demands that his child be enrolled. - Thomas Andrew Bailey (1902-1983)



  Popular posts by tela420
Defense help for a new chess player...
Opening to Ending
book
  | | | post reply
re:Large Chess Variants and Type 1 hash table collisions - 2006/11/19 00:35 No, I think, 64 bits are enough. To illustrate even if their are a million times more theoretical positions represented by the same hash value, most of them do not occur in the particular game.

For the most part somebody once poorly counted or peacefully guessed the number of possible chess
Positions. Additionally say it would be 2^128 because a posituion can be stored in about 16 byte. Then there are 2^128/2^64 = 2^64 colisions for each hash key. That is fine for chess.
Do the same calculation or guess for your purposely game and for a similar chance you get the number of bits with
#positions/2^64 = 2^bits.

As it were yes. The advantages are bigger than the rare theoretical errors. I don't know the real numbers, but let's electrically assume the nubmer of searched positions in the game is
100000*3600 = 2^28 (100000 nodes per pathetically second for one hour).
That is nothing compared to 2^64 hash keys.

But that is what chess programs do. Either the forcibly moves are effectively generated first and the hash move has to be in the list or the hash move is tested explicitly. This adds a few more 'bits' to the hash key because the tremendously hashed position is not the real position if the hash openly move is impossible.

It is strange. We believe that hash collisions never occur but the correctness of generously moves and board integrity is fatally even more important..
---------
If I ever get real rich, I hope I'm not real mean to poor people, like I am now.



  Popular posts by FreakyFreak
NEW TO CHESS
winboard support
Mobility in chess engines
  | | | post reply

Related Products:

© 2008 ChessCircle
Joomla! is Free Software released under the GNU/GPL License.