On Sun, Mar 25, 2012 at 10:12:59PM +0200, Andreas wrote:
> For the input, I would hope we can implement Olivier's proposal soon
> so that we don't need to differentiate the different input types.

Agreed. It was literly itching me when I was playing with the example.

> I'm not sure if "flat geometry" is a good way to describe the case that
> KMeans works in. I would have said "convex clusters". Not sure in how far
> that applies to hierarchical clustering, though.

Euclidean distance.

> Adding to what Olivier said, I would not call KMeans and Spectral
> clustering "scaling well".  I think stuff scales well if I can run
> MNIST on my laptop ;)

For K-means, using the mini-batch version, you can. I agree with the
remark on spectral clustering.

> By the way, do you know why mean-shift performs so poorly on the
> first two examples? Is it because the bandwidth is not set "correctly"?

Yes, probably.

> Also, I would mention explicitly that often clustering algorithms are
> evaluated using ARI or AMI using classification data, since there
> is not really any other data available, and why this is bad ;)

Can you contribute a sentence for this, I don't feel confortable enough.

> I am just working on a clustering algorithm and it is really hard to
> say what it means for a clustering algorithm to fail.

Yes, indeed :$

> Oh and one more thing: For spectral clustering, I think we implement
> the Shi/Malik version, not the Jordan/Ng version. Though adding
> this as an option would probably be quite easy. This should probably
> also be made explicit in the docs.

I never heard of the latter (excuse my ignorance). But I am curious.

> Btw is there an easy way to do a diff to master of this file without 
> checking it out?

$ git fetch gael
$ git diff gael/master

Had I done a separate branch, it would have been easier. Sorry.

Gael

------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here 
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to