Closed caprone closed 5 years ago
The CV.SuperLearner
function is a helper function if you want to estimate the expected performance of the super learner and algorithms in the associated library. CV.SuperLearner
isn't the main function to estimate a super learner predictor, that would be the SuperLearner
function.
CV.SuperLearner
returns V (default is 10) super learner estimates on subsamples of your dataset, but doesn't estimate the super learner on the entire dataset.
Ok thanks; then I can also estimate algorithms's coefficients in Superlearner by cvControl method instead using CV.SuperLearner
Yes, SuperLearner
is the function to use to estimate the coefficients. This function utilizes V-fold cross-validation to estimate the algorithm coefficients and also estimates each individual algorithm on the full provided dataset to be able to construct a super learner predictor.
CV.SuperLearner seems does not return 'superlearner' model, right? (then it's impossible to exploit cv_sl for predictions on new dataset); then, ther's way to pass returned coefficients by CVsuperlearner to SuperLearner?
in this way would avoid a further cross validation in Superlearner to estimate "same" coefficients.
thanks