k-fold Cross Validation in Classification Learner

Hi there, Just wandering how data is partitioned in k-fold cross validation in Classification Learner? Is the data partitioned into k-folds according to class/label or complete random? That is, Is data split up into n subsets according to n classes, partitioned in the class subsets, and then grouped together into the folds used to train/test?

Answer 1

[ahmed nebli]

in k-fold cv, the the data is splited to k-1 for taining and 1 for testing.(e.g.: if you have 100 subject and you use 10-fold cv it would be splited into 90 subject for training and 10 subjects for testing and you make k iteration each time the date is re-splited to 90 and 10.

[James]

Yes but is the data as a whole split in one step, or are the class subsets split and then merged? So in your e.g. you mentioned subjects, which I presume equate to data instances, but each instance is labelled. My question is whether data is split according to labels so for e.g.2 if there are 4 classes with 100 instances each with 25 instances, is the data split as a whole, or is the data split within the class subsets and then merged to maintain the proportionality?