tnagler / kdevine Goto Github PK

This example creates a C stack issue:

library(kdevine, quietly = TRUE)
data(wdbc)  # load data
wdbc[wdbc == 0] <- .Machine$double.eps
fit <- kdevine(as.matrix(wdbc[sample(nrow(wdbc), 2e3, replace = TRUE), 2:3]),
                       xmin = rep(0, 2), cores = 1, info = F)

I tried this both on Mac and Ubuntu. If you reduce the number of samples from 2e3 to 1e3, error goes away. I traced it back to locfit, but not sure how to fix it.

Hi again. I profiled

if (!require(profvis)) install.packages("profvis")
library(kdevine)

data(wdbc)
wdbc[wdbc == 0] <- .Machine$double.eps

profvis::profvis(
  kdevine(as.matrix(wdbc[sample(nrow(wdbc), 2e4, replace = TRUE), 2:6]),
                           xmin = rep(0, 5), cores = 1, info = FALSE)
)

with profvis and have noticed that ~82% of time is spent on garbage collection. For a dataset of ~2 Mb, ~100 Gb worth of objects was allocated and destroyed.

Do you have any insight into why that is happening and whether any fixes come to mind?

Dear Nagler

I need help in R, please.
usually, any package of any distribution in R contains three commands such as rname, dname and pname. I found a package called kdevine https://github.com/tnagler/kdevine

and it has dkdevinecop and rkdevinecop but it doesn't have the option pkdevinecop (the CDF). I tried to write it like this but it is wrong, could you see it, please

Regards

library(kdevine)
library(kdecopula)

data(wdbc)

fit <- kdevine(wdbc[, 5:7], xmin = rep(0, 3))
f<-dkdevine(wdbc[, 5:7], fit)

for(i in 1:length(f)){
p<-sum(dkdevinecop(c(wdbc[, 5],wdbc[,6],wdbc[,7]),fit))
}
print (p)

Hey again :) One issue I am running into is handling categorical variables in a principled manner - I jittered mine to get an approximate answer.

Are there any theoretical obstacles with adding something like Aitchison-Aitken kernel and the corresponding CDF to also handle unordered, categorical data, similar to np package, while still maintaining control of the curse of dimensionality?