back-projection analysis

Fixed our pool cpc... we orginally have a dividing problem when we were pooling the cov...

Run newphil.R to aviod using data... only require covs and npts to run.

Phillips pool eigen values are not ordered (bp problem if highest eigenvalue is interested because our bp code takes first eigen vector as default) This does not concern us at all because our pool cpc is ""correct"" (consistent with eigen value ordering)

test_examples

Bank data: - full cpc ok -(OUR) partial cpc does not select the highest eigenvalue implies the constrained eigen value is not right.

Iris data: -full cpc ok -partial cpc a little off + 3rd cov one sign off (Phillips) (2 out of 4 cpc) -partial cpc does not order after the first one for one common pc (1 out of 4) diddo

Marten data: -full cpc ok (eval ordering is kind of off, resulting in minor variations in numbers. BUT (not 100% sure), it SHOULD not affect the estimated population cov because it is being multiplied with the corresponding eigen vaules and vectors.

-partial cpc does not select the highest eigenvalue... it matters because it affacts which eigenvector constrain to be the same (1 out of 4) diddo

-partial cpc (2 out of 4) diddo

Turtle data: -full cpc ok -partial cpc ok but numbers are a little off

Vole data: -full cpc ok.. a little off - partial cpc ok ...

library(cpcbp)
source("newphil.R")
source("mikecpc.R")
source("test_examples.R")

newphil(x,test_bank,n_bank)$datlist[[3]]
mikecpc(test_bank,n_bank,1)

newphil(x,test_marten,n_marten)$datlist[[2]]
mikecpc(test_marten,n_bank,2)