How does the classification learner app implement k-fold cross validation?

Hi

Does anyone know how the k-fold cross validation is implemented in the classification learner app? Specifically, does it automatically stratify the folds?

Thanks

James

Answer 1

[the cyclist]

I am most definitely not an expert in this app, but I built a simple model using the default settings, and then exported the function. It has this line

% Perform cross-validation
partitionedModel = crossval(trainedClassifier.ClassificationTree, 'KFold', 5);

which suggests to me that it does not stratify by default. (The crossval function accepts a 'Stratify' Name-Value pair, which is not present here.)

I could not see a way to specify this in the app itself, but maybe there is. You could contact support to find out, if you don't get an answer here.

[James Alix]

Ok, thats a good idea.

Searching around today I found this:

By default, crossval ensures that the class proportions in each fold remain approximately the same as the class proportions in the response variable ... on this cvpartion page (https://uk.mathworks.com/help/stats/cvpartition.html#d124e288211) but I'll email support and see if I can confirm.

Thanks

James

[the cyclist]

Ah, I see that whether crossval stratifies by default or not is dependent on the nature of the first argument (just a number, or the group).