GMM LDA

[Thomas Richardson]

Hi all,

So i have a dataset of 688 faces and a binary grouping variable (let's call it 2 types of photo) and want to run LDA on it. I have 53 landmarks so my procrustes aligned dataset is a 53x2x688 matrix. As they are faces they have been aligned assuming object symmetry, so i think the "shape" variable in the geomorph dataframe reflects the symmetrical component of variation. 

I heard that i need to convert it to a 688x106 matrix using two.d.array, but when i do that and run the LDA i get :

"Error in lda.default(x, grouping, ...) :  variables  38  72  78  92 102 appear to be constant within groups"

This presumably means that these variables have no variance within the groups on these landmarks. However when i remove these variables it just says the different ones are constant, and it finds a new one whenever i remove one! 

Does anyone know what's going on here? i've been at this for like a day now.

Cheers,

Richie

[Mike Collyer]

Hi Richie,

I would use plotTangentSpace on your shape variables instead of two.d.array.  If you do not mess with the default arguments, the $pc.scores object created will shave off the extra dimensions.  If you do this and still have trouble (you shouldn’t), you can always remove the last column or two of the matrix, as the principal components will have the most variation among specimens represented in the earliest dimensions.

I was not aware that lda.defult had such a trap.  It would not be an issue for classification, but maybe it’s there to prevent strange discriminant functions.  Hopefully using PC scores will not disrupt your intended use of discriminant analysis.

Cheers!

Mike

--
You received this message because you are subscribed to the Google Groups "geomorph R package" group.
To unsubscribe from this group and stop receiving emails from it, send an email to geomorph-r-pack...@googlegroups.com.
To post to this group, send email to geomorph-...@googlegroups.com.
Visit this group at https://groups.google.com/group/geomorph-r-package.
To view this discussion on the web, visit https://groups.google.com/d/msgid/geomorph-r-package/867b50c3-4a3b-4545-bf78-4164acee8e30%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Thomas Richardson]

Fantastic, that worked perfectly!

Cheers,

Richie

On Thursday, November 23, 2017 at 2:03:21 PM UTC, Michael Collyer wrote:

Hi Richie,

I would use plotTangentSpace on your shape variables instead of two.d.array.  If you do not mess with the default arguments, the $pc.scores object created will shave off the extra dimensions.  If you do this and still have trouble (you shouldn’t), you can always remove the last column or two of the matrix, as the principal components will have the most variation among specimens represented in the earliest dimensions.

I was not aware that lda.defult had such a trap.  It would not be an issue for classification, but maybe it’s there to prevent strange discriminant functions.  Hopefully using PC scores will not disrupt your intended use of discriminant analysis.

Cheers!

Mike

On Nov 23, 2017, at 6:49 AM, Thomas Richardson <thomasricha...@outlook.com> wrote:

Hi all,

So i have a dataset of 688 faces and a binary grouping variable (let's call it 2 types of photo) and want to run LDA on it. I have 53 landmarks so my procrustes aligned dataset is a 53x2x688 matrix. As they are faces they have been aligned assuming object symmetry, so i think the "shape" variable in the geomorph dataframe reflects the symmetrical component of variation. 

I heard that i need to convert it to a 688x106 matrix using two.d.array, but when i do that and run the LDA i get :

"Error in lda.default(x, grouping, ...) :  variables  38  72  78  92 102 appear to be constant within groups"

This presumably means that these variables have no variance within the groups on these landmarks. However when i remove these variables it just says the different ones are constant, and it finds a new one whenever i remove one! 

Does anyone know what's going on here? i've been at this for like a day now.

Cheers,

Richie

--
You received this message because you are subscribed to the Google Groups "geomorph R package" group.

To unsubscribe from this group and stop receiving emails from it, send an email to geomorph-r-package+unsub...@googlegroups.com.