Bubble maps
M Joonas Pihlaja
Hi all,
Due to popular demand I've created Self-Organising Map
visualisations of the warriors on Koenigstuhl. You can find
them in Postscript at:
http://www.cs.helsinki.fi/~jpihlaja/cw/maps/
[I'm not converting them to .pdf as they come out much too
large.]
Thanks to Christoph for making the score data of Koenigstuhl
available.
SOM is a method for mapping high-dimensional data sets that
groups like elements together. You can find out more about SOM
here:
http://www.cis.hut.fi/~hynde/lvq/
http://www.cis.hut.fi/research/som-research/nnrc-programs.shtm
Enjoy,
Joonas
M Joonas Pihlaja
Oh my god, I'm becoming senile: Just now I meant to correct the
mistakes in the post (added below) and replied directly to the
author! :-}
From jpih...@cc.helsinki.fi Tue Sep 3 21:41:42 2002
Date: Tue, 3 Sep 2002 21:40:54 +0300 (EET DST)
From: M Joonas Pihlaja <jpih...@cc.helsinki.fi>
To: M Joonas Pihlaja <jpih...@cc.helsinki.fi>
Subject: Re: Bubble maps
On Tue, 3 Sep 2002, M Joonas Pihlaja wrote:
> Due to popular demand
Okay, okay,... just the one, but he was insistent.
[snip]
> [I'm not converting them to .pdf as they come out much too
> large.]
I meant to deny conversion to .png.
Joonas
Philip Thorne
> Hi all,
>
> Due to popular demand I've created Self-Organising Map
> visualisations of the warriors on Koenigstuhl. You can find
> them in Postscript at:
>
> http://www.cs.helsinki.fi/~jpihlaja/cw/maps/
>
> http://www.cis.hut.fi/~hynde/lvq/
> http://www.cis.hut.fi/research/som-research/nnrc-programs.shtm
> [...]
>
> Joonas
Interesting. Above links aside, I tried to find some simple lit.
on the topic but it's getting late and the books seem quite pricey
so here you find me attempting the lazy route to enlightenment...
I found the explanation you made to a similar posting in the past which
helps greatly but judging by the placement of Fire & Ice II and Damage
Inflicted (X map) the pattern / layout isn't the same as then. Is this
something that arises out of the mapping process/algorithm or do you have
some control over the groupings, score excepted? For example, last time
you commented on a progression by age - I'm assuming that was solely
the result of newer warriors generally performing better than older ones
rather than age being an input.
[If you can formulate the answer in such way that Slimy Time Vamp looks
good so much the better <g>]
Something else that intrigues me is the shading given to unoccupied
hexagons - e.g. pure white under Nova-A.1b.
Lastly, if I'd noticed your earlier posting I too would have encouraged
the follow-up. tx [looking at pictures is so much easier than coding...]
Philb
M Joonas Pihlaja
On 3 Sep 2002, Philip Thorne wrote:
[snip]
> Interesting. Above links aside, I tried to find some simple lit.
> on the topic but it's getting late and the books seem quite pricey
> so here you find me attempting the lazy route to enlightenment...
Stick in "Self Organizing Map" into google and read on.
Somebody's thesis (with a good description of the algorithm):
http://www.cis.hut.fi/~tho/thesis/
A Very Brief Summary of The Self-Organizing Map
http://www.canis.uiuc.edu/~dpape/dlisom/summary.html
The idea is that you start out with bunch of vectors, the
"samples", in a space that are drawn from some unknown
distribution and you want to "cover" or "fill" the space with a
rubber sheet so that it somehow reflects the unknown
distribution.
> I found the explanation you made to a similar posting in the
> past which helps greatly but judging by the placement of Fire
> & Ice II and Damage Inflicted (X map) the pattern / layout
> isn't the same as then. Is this something that arises out of
> the mapping process/algorithm or do you have some control
> over the groupings, score excepted?
I don't impose any constraints on the grouping. The algorithm
groups them all by itself from (only) the score results.
> For example, last time you commented on a progression by age
> - I'm assuming that was solely the result of newer warriors
> generally performing better than older ones rather than age
> being an input.
It was astonishing to me that SOM displayed concretely exactly
the kind of distinguishing power that it is claimed to have. That
warriors from the same period are grouped together reflects the
environment they were created in, and shows how large the gap is
between 'old', and 'new' strategies.
For example, in the -88 map we can clearly see that there was a
time when imp-only strategies were all the rage, but today it is
inconceivable that a warrior that relies only on imps for its
attack and defence could get anywhere near a hill. That the same
grouping-by-age phenomenon doesn't appear on the X map, I
suspect, is due to the small number of warriors on that hill.
Maybe it means that the X environment is underexplored?
> Something else that intrigues me is the shading given to unoccupied
> hexagons - e.g. pure white under Nova-A.1b.
The shade indicates the density of sample vectors that a hexagon
thinks it covers, with white being more dense. A round dot
inside a hexagon means that no sample vectors are covered by it.
> Lastly, if I'd noticed your earlier posting I too would have
> encouraged the follow-up. tx [looking at pictures is so much
> easier than coding...]
Thanks, glad you enjoyed it.
Joonas
met...@uboot.com
Only just found these, been offline.
I've had a quick glance, they look extremely
interesting and I will be having a closer look at
first opportunity. Interesting also to see how
some of the warriors have been grouped.
Pity I don't have an A3 (or larger) printer - my
print-outs don't really do them justice.
Thanks & regards,
John
--
SMS Me on +44 7866 049 839
_____________________________________________________________
300 character txt msgs, ringtones, logos & much, much more...
http://www.uboot.com - who r U?