[LRUG] Battleships last night

[Tim Cowlishaw]

On Tue, Oct 11, 2011 at 9:48 AM, Murray Steele <murray.steele at gmail.com> wrote:
> Hi all,
> Even though my own Rear Admiral h-lame was knocked out in the first round I
> really enjoyed working on my bot and seeing what everyone else had come up
> with.  To that end, I wonder if Paul could put all the contestants on github
> so I can poke around and see why mine was so crap?

+1 - I'd be very curious to see how mine would have fared without
deploying the unintentional 'torpedo to the engine room' tactic that
it was using on the night!

> Watching the games last night I was struck that most people had gone for a
> hunt vs. kill mode.  The hunt modes seemed mostly to diagonally stripe the
> board until a hit is made.  The kill modes seemed mostly to circle the hit
> looking for other hits and then track along that "hit path" till it ran out
> of hits.  Anyone implement their hunt vs. kill modes in an interesting way?
> I also wonder if anyone else took an approach that was markedly different
> from this?

I didn't go for seperate hunt and kill modes, instead choosing a
random square from within a probability distribution that is uniform
at the beginning of the game, and is updated with each hit or miss to
reflect the increased or decreased likelihood of a ship being in the
surrounding squares. However, this approach remains untested until I
can actually get the thing running :-)

> Finally, did anyone do anything interesting with ship placement?  I saw at
> least one bot was only placing the boats on the edges, and conversely one
> bot that would never place boats on the edges, and I'm sure there was at
> least one bot that always used the same boat placement.

This is a really interesting question - I worked on the assumption
that ships in the middle of the board are harder to find than ships at
the edge (as being at the edge limits the number of different
directions that a ship could sit in), but am still unsure about
whether placing ships adjacent to each other offers any advantage or
disadvantage. Assuming no playing strategy retains state between
games, I can't see why using the same placement each time would
necessarily be a disadvantage, however that raises another interesting
question - did anyone's bot learn from previous games in order to
perform better in subsequent ones?

Also, would anyone be up for a no-holds-barred "bareknuckle
battleships" bout (potentially at LRUG nights?) where no techniques
are off limits, and the only objective is to scupper your opponent's
bot through the most nefarious means possible? :-)


Tim