Hi team FLU_ATTACK. I have some questions about your permutation experiment. Did you do LOOCV for computing predictions on train set or just trained your model on permuted data and predicted on original train set? Also did you generate permutations by swapping the rows or used a built-in function like randpermute in python? Thanks.
Created by Zafer Aydin zaferaydin Hi, sorry about the delay. So, we did permutations just for the second subchallenge. In this, we were working with a random forest algorithm, so the estimate for the training error actually comes automatically from the way the random forest method works (leaving out a random set of individuals in each tree - which can be "tested" as if they are from the test set). LOOCV is not necessary here.
As for the other question: the permutations were generated using the "sample" function in R. Cheers