Hi Michel,
Pretty useful project you have here, I haven't read the
clustering algorithms yet but have used it (with
ecological similarity indexes among samples!).
Hierarchical clustering makes that a LOT more useful,
so thank you very much :)
This patch adds a topological output to
Cluster(.topology()) and a respective method to
BaseClusterMethod(.topo()). It also changes .data()
behavior to avoid the .data()[0] idiom.
Feeding the output of .topo() to
http://aspn.activestate.com/ASPN/Cookbook/Python/Recipe/139422
gives (e.g.):
9 -----+
|--+
8 -----+ |
|--
# --+ |
|--+ |
w --+ | |
|--+
e -----+
Regarding the .data() change:
* I think it'd be better to keep two different
datasets (.raw and .data?), because both states (flat
and clustered) are pretty different beasts.
* Perhaps having .data() "contents" stored as
.data would make things easier (in a "why call a method
with no parameters, that returns an object" way :)).
Making .data a property would keep current behavior and
require less ()s :)
I'm working on an advanced version of the topological
view and on a derivative of that printing recipe for my
own needs, but will forward them here when done.
Thanks again,
Daniel
Topology patch for cluster.py
Logged In: YES
user_id=1200609
Duh, of course SF mangles the ASCII art.
Here's something I'm getting from very ugly, very poor code.
It's based on the recipe above, plus handling of tags
(levels) and a hack that lets Cluster.topology() know that I
want a label instead of each item's repr.
ASCII art for clusters.
Logged In: YES
user_id=560690
Hi Daniel,
thanks for the input. It gives me the warm and fuzzy
feeling, that this project is indeed still in use ;)
Having topological output for the clusters is a brilliant
idea. I remember well, a couple of weeks ago. I was sitting
at the desk with several A4 pages taped together, and was
drawing the output of python-cluster for debugging. Hehe...
;)
Having this is very useful. I had a quick look through your
code, and it looks fine. I will merge it as soon as I get
the time.
Thanks a lot,
Michel
Logged In: YES
user_id=560690
Bah, I accidentally updated an old revision with this
patch. Need to re-do the work :(