Re: PCA - separation and variance
- From: Paige Miller <pmiller5NOSPAM@xxxxxxxxxxxxxxxx>
- Date: Sat, 08 Jul 2006 22:19:30 GMT
On 7/3/2006 4:17 AM, shay@xxxxxxxxxxxxxxx wrote:
I have a multi-dimensional data set containing two different groups of
data I want to separate utilizing PCA. I observed (by eye) that a
certain set of variables gives me a good separation using the first
three PCs.
You are using fundamentally the wrong method. Use discriminant analysis here instead of PCA. Then you won't have to determine by eye which variables give you a good separation. You can use statistical methods and tests to determine which variables give you a good separation.
What is the correlation, if it at all exists, between the separation inSince PCA is the wrong method to use here, i.e., it optimizes a criterion that can be unrelated to separation between groups, I would guess that there is no relationship between separation in 3D and separation in higher dimensions.
3D to the actual separation in the higher dimensioned plane? How does
the variance of the solution come into play? Can I say that the
separation on the higher dimensioned plane is as good as or perhaps
better/worse than the one in the 3D plane?
--
Paige Miller
pmiller5@xxxxxxxxxxxxxxxx
It's nothing until I call it -- Bill Klem, NL Umpire
If you get the choice to sit it out or dance,
I hope you dance -- Lee Ann Womack
.
- References:
- PCA - separation and variance
- From: shay
- PCA - separation and variance
- Prev by Date: Re: version of Golub Chan LeVeque stable algorithm for sample variance applicable to weighted observations
- Next by Date: Re: version of Golub Chan LeVeque stable algorithm for sample variance applicable to weighted observations
- Previous by thread: PCA - separation and variance
- Next by thread: question about training samples for logit
- Index(es):
Relevant Pages
|
