[Melbourne-pm] Timer::HiRes and alarms? (Was: AI Contest)

[Nathan Bailey]

On 11/11/2010, at 5:46 PM, Toby Corkindale wrote:
> The "local $| = 1;" was inherited from the original Perl starter  
> package, which is of dubious quality; it might be better to make it  
> "$| = 1;" instead. ie. Global, not local.. In case the alarm signal  
> handler is occurring inside a different scope?
> (That's clutching at straws, but worth a shot..)

Ahh, good point - the signals may be happening in some other scope.

> Likewise, I've found my code always has plenty of time to spare,  
> apart from poor coding.. but then, my bot isn't that complicated  
> either.
> (He just runs through three phases - 1) global analysis; 2) per- 
> planet defensive analysis; 3) per-planet further analysis and moves.
> It's more like an interconnected heuristic system, rather than  
> proper AI. Although I *want* to make it have more I and less A than  
> it does currently.. when time allows :/ )

Despite having studied AI and neural nets, mine isn't that different.  
Actually, this is a problem that lends itself very well to neural nets  
(= finding out the optimal value for each target planet). But it's  
been a while since I've done either - I don't even get to code perl  
much these days :-)

> Hmm, there's already source for the C and Java implementations of  
> the engines. If you're worried about your implementation, maybe it  
> would be better to just modify one of the stable engines to allow  
> for more debug output?

Cool, I might play with the C one. The (compiled) java version in your  
repository doesn't provide any feedback, so when I crash/timeout, I  
don't know where or why.

> Yeah, it could definitely do with some improvements.
> My Git repo holds a few by now, but only so far as improving the  
> existing methods.

Nice! What does this line do (Planet.pm and Fleet.pm)?
	use overload '""' => sub { shift->PlanetID };

Is there a reason you don't cache the distances (eg. in a hash?) -  
although I tried it, and NYTProf told me it was worse! :P

I also wonder how evil it is to dig into the object, eg. for GetPlanet:
	return ${$self->{_planets}}[$planet_id];
(Okay, I know how evil it is, I just wonder if it's an appropriate  
shortcut given the amount of time that code runs)

> I've added various things (privately) such as locating nearby  
> planets, summing fleet-effects and predicted planet populations, but  
> kept them private. The published code is meant to just be a starter  
> package, after all.
> .. plus my code isn't very good yet, anyway.

*nod* I think strategy-oriented sub-routines should be private, but I  
think closest planet is pretty global, ie. return either:
	a) a list of planets in distance order
	b) a list/hash of lists of planets indexed by distance

> That does sound odd; something must be quite broken if it's failing  
> at that basic level of I/O! Can you write a simple test harness to  
> operate on the methods parsing the game-state only?


Do you mean the 'while(1) { my $current_line = <STDIN> ... }' bit or  
my own code in DoTurn()?
N