Classify.bgPCA() does not give correct results

[IrisMenendez]

Hi, first thanks for implementing between-group PCA in your package. I'm using it to classify fossil specimens in dietary categories based on the teeth shape of extant specimens. I have all shapes (extant and fossils) superimposed in a gpagen objectGPA.all and a factor with the diet of extant species data_inf$DietCVA. Then I use this code:

bgPCA <- groupPCA(GPA.all$coords[,,1:1101], groups= data_inf$DietCVA, cv=T)
class_bgpca <- classify(bgPCA, newdata= two.d.array(GPA.all$coords[,,1102:1222]), cv=T)

I obtain that all the specimens are insectivores, which is not possible because is the category less probable (also see figure: pink filled dots are fossil specimens plotted in the groupPCA space using predict()).

> class_bgpca
specimens have been classified as:

 Insectivore 
        121 

This does not change when I specify the prior probabilities. I don't know if it's my fault or a bug in the function. Maybe I'm using the wrong object for the newdata?

Thank you, Iris

[zarquon42b]

Hi Iris, is this intentional that you use other indices for predicting? Also, two.d.arry arranges the coordinates in another way as Morpho. Try vecx instead and check back. class_bgpca <- classify(bgPCA, newdata= vecx(GPA.all$coords[,,1102:1222]), cv=T)

However, I just adapted the code so this will be handled automatically. If you install the latest snapshot from github using devtools, you can simply run: class_bgpca <- classify(bgPCA, newdata= GPA.all$coords[,,1102:1222], cv=T)

Best

Stefan

[IrisMenendez]

Thanks, Stefan! It worked perfectly. I will keep that in mind for the future.

I'm not sure what you mean by using different indices for predicting, I think I used the same ones. The coding was

predict_bgpca <- predict(bgPCA, GPA.all$coords[,,1102:1222])

#plotting
f_s <- ggplot(data=as.data.frame(bgPCA$Scores),aes(x=V1,y=V2))
f_s + 
  geom_point(aes(col=data_inf$DietCVA), alpha = 4/10, shape=21, stroke=1, size=2) + 
  scale_color_manual(values=col.diet)+
  geom_point(data=as.data.frame(predict_bgpca), aes(colour=class_bgpca2$class), shape=16, size=2)+
  ggtitle("landmarks bgPCA")+
  theme_classic()

Thanks a lot! Iris

[zarquon42b]

Sorry, my bad, everything is all right, I just realized that it wasn't the landmark indices but the specimen indices (I was a bit tired when I wrote the reply).