--- Mark Boon <[EMAIL PROTECTED]> wrote:
>
> On 29-mrt-08, at 10:47, Jacques Basaldúa wrote:
>
> > I don't use don't care. I mask code in 2 bits:
> void, own,
> > opponent, out_of_board. This produces bigger
> database size,
> > because the only way to introduce "don't care" is
> to repeat the
> > pattern.
>
> OK, so we were comparing apples and oranges. I know
> that using a hash-
> code with fixed-size patterns is a whole lot faster.
> You don't need
> to revert to assembler to make that really fast. If
> you see my
> original message you'll see that I mentioned that. I
> was specifically
> asking about a pattern-matcher that allows for
> don't-care points
> because I think I need that.
>
> It's possible that speed considerations will make it
> necessary to use
> this kind of fixed-size patterns. I'm not yet
> convinced of it, but I
> don't rule out the possibility either. But before
> turning to the dark
> side I'd prefer to first explore how far (and fast)
> I can get with a
> 'traditional' pattern-matcher that allows for don't
> care points.
Considering how inexpensive memory is, and how
branches cause processor pipelines to stall, it seems
to make sense to convert "don't care" patterns into
however many fixed patterns would be equivalent. If
there are three "don't care elements", which could be
instantiated three ways, then 3^3 patterns would
replace one, for example. If these were converted into
optimized assembler, per Jacques' suggestion, the
pattern matcher would be extremely fast.
Terry McIntyre <[EMAIL PROTECTED]>
Wherever is found what is called a paternal government, there is found state
education. It has been discovered that the best way to insure implicit
obedience is to commence tyranny in the nursery.
Benjamin Disraeli, Speech in the House of Commons [June 15, 1874]
____________________________________________________________________________________
Like movies? Here's a limited-time offer: Blockbuster Total Access for one
month at no cost.
http://tc.deals.yahoo.com/tc/blockbuster/text4.com
_______________________________________________
computer-go mailing list
computer-go@computer-go.org
http://www.computer-go.org/mailman/listinfo/computer-go/
