I think there are three ways the tree is stored,

 

As a tree, with pointers

As a tree (with pointers), and with a separate transposition table - I think
Fuego does this

As a hash table only, with no pointers between nodes.  This is what Many
Faces does.

 

I never implemented a tree structure, so I can't compare the strength of the
tree to the strength of a table only.  For me, the hash table structure
seems easier to code, since there are no pointers.  It's far easier to
garbage collect.

 

David

 

 

From: [email protected]
[mailto:[email protected]] On Behalf Of Don Dailey
Sent: Thursday, May 06, 2010 11:07 AM
To: [email protected]
Subject: Re: [Computer-go] effectiveness of transposition tables for go

 

When you say transposition table I'm not sure what you mean.   Most programs
do not do a typical alpha/beta search where transposition tables are used.

In MCTS there seems to be 2 different ways the "tree" is stored in memory,
via a transposition table or via a tree structure (where transposition would
not be noticed.)      Is that what you are  talking about?

In the old Lazarus program, which is pretty weak by todays standards, I used
a tree because it seemed a bit easier to implement and I like easy.    I did
not want to deal with how to handle transpositions such as how to identify
the parent node when there might be many and how to reconcile different hit
counts and success rates.  It's probably not a difficult thing, but I wanted
to keep it real straightforward.  

But I think many excellent programs use the hash table structure which I
would call a transposition table.   Is that what you mean?    My intuition
is that a transposition table should be superior.  

Don




On Thu, May 6, 2010 at 1:45 PM, Jean-loup Gailly <[email protected]> wrote:

Petr Baudis and I have a debate about the effectiveness of transposition

tables for go.  My guess is that, compared to a simple tree, a transposition

table should make the program stronger, particularly for long games and many

cores.  Petr thinks that transposition tables are much less effective in go

than in chess.  For those of you using a transposition table, did you

observe a strength increase? If so do you have even a vague estimate of how

many elo points this can bring? Is it more effective in 19x19 or 9x9?

Thanks.

 

Jean-loup

 


_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

 

_______________________________________________
Computer-go mailing list
[email protected]
http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

Reply via email to