| Robust covariance estimate [message #94289] |
Tue, 28 March 2017 12:19  |
natha
Messages: 482
Registered: October 2007
|
Senior Member |
|
|
Hi all,
I am performing a Principal Component Analysis and I would like to use robust computation of the covariance matrix to avoid the outliers affect my results.
I've been reading a little bit about it and it seems that there are many approaches to compute robust covariance estimators.
I didn't find any code for that, only a a Python library which requires the input data matrix to be Gaussian distributed, which is not my case:
http://scikit-learn.org/stable/auto_examples/covariance/plot _robust_vs_empirical_covariance.html
Any ideas or suggestions?
Thanks in advance for your help,
nata
|
|
|
|
| Re: Robust covariance estimate [message #94290 is a reply to message #94289] |
Tue, 28 March 2017 13:04  |
wlandsman
Messages: 743
Registered: June 2000
|
Senior Member |
|
|
I'm not much help except to say that the requirement for a Gaussian distribution is not very strict for most robust statistics programs (e.g.
https://idlastro.gsfc.nasa.gov/ftp/pro/robust/). Presumably, you have a centrally concentrated distribution - otherwise how could one identify outliers? Assuming a Gaussian for the sole purpose of identifying outliers might be adequate. (If not Gaussian, do you know the true distribution of your data matrix?) -Wayne
On Tuesday, March 28, 2017 at 3:19:51 PM UTC-4, nata wrote:
> Hi all,
>
> I am performing a Principal Component Analysis and I would like to use robust computation of the covariance matrix to avoid the outliers affect my results.
> I've been reading a little bit about it and it seems that there are many approaches to compute robust covariance estimators.
> I didn't find any code for that, only a a Python library which requires the input data matrix to be Gaussian distributed, which is not my case:
> http://scikit-learn.org/stable/auto_examples/covariance/plot _robust_vs_empirical_covariance.html
>
> Any ideas or suggestions?
> Thanks in advance for your help,
> nata
|
|
|
|
comp.lang.idl-pvwave archive
Members
Search
Help
Login
Home
