number of permutations in permimp::permimp function

Question

I am using permimp function from permimp package. What value shall I assign to nperm? 100? or 10 will be enough? My dataset contains 35 variables and around 800 observations

crf <- party::cforest(response ~ ., 
                     data = mydata, 
                     control = cforest_unbiased(mtry = 3, ntree = 1500))

cf_vi <- permimp::permimp(crf, conditional = TRUE, nperm = 100)

Also where could I look for examples of using cforest/permimp?

Answer 1

The number of permutations should be large enough so that the results do not depend much on the exact number of permutations or the setting of the random seed. With just 10 permutations the results probably still differ qualitatively when you run the same code with different seeds.