[scikit-learn] Markov Clustering?
Allan Visochek
avisochek3 at gmail.com
Tue Dec 6 08:05:21 EST 2016
At it's core, Markov clustering is a graph algorithm, it operates on a
sparse similarity matrix (essentially, by simulating flow between the data
points). This makes it useful for similarity graphs that don't originate
from features (i.e. protien-protien interaction networks). Because the
graph is based on similarity though, its definitely possible to use it as a
data clustering algorithm that takes a similarity metric as an argument.
I suppose it could be implemented so that the algorithm could take either a
sparse similarity matrix or a set of features as its first argument. This
would keep the same structure of the other clustering algorithms, but also
allow use with pure similarity graphs. Does this make sense?
On Mon, Dec 5, 2016 at 8:57 AM, Andy <t3kcit at gmail.com> wrote:
>
>
> On 12/05/2016 08:51 AM, Raphael C wrote:
>
>> And...
>>
>> [1] Stijn van Dongen. Graph Clustering by Flow Simulation. PhD
>> thesis, University of Utrecht, May 2000.
>> http://www.library.uu.nl/digiarchief/dip/diss/1895620/inhoud.htm
>>
>> has
>>
>> 1201 citations.
>>
>> I think it's fair to say the method is very widely known and used.
>>
>> Ok cool.
> I haven't looked at it, my question is now whether this is more of a
> "graph clustering"
> or a "data clustering" approach, though that distinction is not very clear.
> Some of the papers compare it against affinity propagation, which we do
> have implemented.
>
> If this algorithm makes sense for knn graphs or similar methods we
> implemented in SpectralClustering,
> then I guess go for it?
> _______________________________________________
> scikit-learn mailing list
> scikit-learn at python.org
> https://mail.python.org/mailman/listinfo/scikit-learn
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.python.org/pipermail/scikit-learn/attachments/20161206/20b3d376/attachment.html>
More information about the scikit-learn
mailing list