Comments on: Dimensionality reduction: comparison of different methods

By: Matt

Matt — Wed, 03 Mar 2010 15:45:51 +0000

Hi serra,

I don’t understand your first remark. I didn’t do a classification in this example, I only compare the distances in the original space to the distances in the reduced space. This is what I display inside the table (percentage of difference between original distances and distances in the reduced space). Classification can be done in this space, but it’s an other subject
The algorithms that use an approximation of the geodesic distance are:

Isomap
Sp
Ssam
Scca

By: serra

serra — Wed, 03 Mar 2010 12:51:27 +0000

Hi Matt,
The result of your test is very important for my research aswell. But I found it a bit difficult to understand. For example if you have done a classification with this reduced space and give the overall classification results, I would have understood better:) Now sorry but, which method is using Geodesic distances (I only know ISODATA) nad what are the results in the table? If you give further information, I would be happy to read it:) Thanks in advance
Regards,

By: Matt

Matt — Wed, 18 Jun 2008 05:54:29 +0000

Hi !

I will post another blog ticket in a few days (after my small holidays), because 99% of the code is already available as a subpart of a scikit. So you can use my code, give me some feedback, … before I make the official announcement.
There is a small tutorial here: http://scipy.org/scipy/scikits/wiki/MachineLearning/ManifoldLearning

Thanks for the comment !

By: lousjiang

lousjiang — Wed, 18 Jun 2008 03:12:37 +0000

Thank you for your article and I find it of great value in my current research area, but I am new in this field,
so I am wondering if you can sent a copy of the codes about this
article(Dimensionality reduction: comparison of different methods), I really appreciate that.
and here is my Email: lousongjiang@hotmail.com, thank you