[Computer-go] Rules for the Computer Olympiad -- Remote Play

Nick Wedd nick at maproom.co.uk
Mon Sep 19 12:13:35 PDT 2011


On 19/09/2011 19:21, Petr Baudis wrote:
> On Sat, Sep 17, 2011 at 11:33:20AM -0400, steve uurtamo wrote:
>>> I guess the question is what do we want to
>>> identify: the strongest program or the strongest artificial playing
>>> entity?
>>
>> I'm starting to be convinced that there's very little difference. Code
>> isn't generally separable from its hardware, once it has been heavily
>> optimized. This has been discussed on the list before, but I'll just
>> point out that if you want to have a competition between *algorithms*,
>> you'd need a very artificial environment that you might have
>> difficulty getting everyone to agree upon.
>
> I'd like to chip in with a concrete example - relative strength of Fuego
> vs. Pachi seems to heavily depend on the amount of computing power.
> On single CPU, Fuego is much stronger than Pachi, they get about even
> with 4 CPUs and from there on, Pachi scales better and quickly gets
> significantly stronger.  (Details to follow in our ACG13 paper.)
>
> Of course we could spend a lot of effort on tweaking Pachi to play
> well on lower-end platforms, but I'd rather be looking for the next
> algorithmic breakthrough and be ready for the many-core computers
> of tomorrow. We should think what do we want to measure in the contest.
>
> If no remote play is allowed anyway, I think the playing field should
> be at least set really even by all players using the same hardware
> platform, e.g. like it was done in Tampere. Restricting remote play
> but still allowing for differentiation based on whatever laptop one
> brings in does not make too much sense to me.

My recollection of Tampere is that standard hardware was provided, but 
entrants could bring their own hardware if they wanted.  Indeed one 
entrant did so.

I am not saying that this is a better way, just that it was how we did 
it in Tampere.

Nick
-- 
Nick Wedd
nick at maproom.co.uk



More information about the Computer-go mailing list