recombineSL: Recombine a SuperLearner fit using a new metalearning method

Description

The recombineSL function takes an existing SuperLearner fit and a new metalearning method and returns a new SuperLearner fit with updated base learner weights.

Usage

recombineSL(object, Y, method = "method.NNloglik", verbose = FALSE)

Arguments

object

Fitted object from SuperLearner.

The outcome in the training data set. Must be a numeric vector.

method

A list (or a function to create a list) containing details on estimating the coefficients for the super learner and the model to combine the individual algorithms in the library. See ?method.template for details. Currently, the built in options are either "method.NNLS" (the default), "method.NNLS2", "method.NNloglik", "method.CC_LS", "method.CC_nloglik", or "method.AUC". NNLS and NNLS2 are non-negative least squares based on the Lawson-Hanson algorithm and the dual method of Goldfarb and Idnani, respectively. NNLS and NNLS2 will work for both gaussian and binomial outcomes. NNloglik is a non-negative binomial likelihood maximization using the BFGS quasi-Newton optimization method. NN* methods are normalized so weights sum to one. CC_LS uses Goldfarb and Idnani's quadratic programming algorithm to calculate the best convex combination of weights to minimize the squared error loss. CC_nloglik calculates the convex combination of weights that minimize the negative binomial log likelihood on the logistic scale using the sequential quadratic programming algorithm. AUC, which only works for binary outcomes, uses the Nelder-Mead method via the optim function to minimize rank loss (equivalent to maximizing AUC).

verbose

logical; TRUE for printing progress during the computation (helpful for debugging).

Value

call: The matched call.
libraryNames: A character vector with the names of the algorithms in the library. The format is 'predictionAlgorithm_screeningAlgorithm' with '_All' used to denote the prediction algorithm run on all variables in X.
SL.library: Returns SL.library in the same format as the argument with the same name above.
SL.predict: The predicted values from the super learner for the rows in newX.
coef: Coefficients for the super learner.
library.predict: A matrix with the predicted values from each algorithm in SL.library for the rows in newX.
Z: The Z matrix (the cross-validated predicted values for each algorithm in SL.library).
cvRisk: A numeric vector with the V-fold cross-validated risk estimate for each algorithm in SL.library. Note that this does not contain the CV risk estimate for the SuperLearner, only the individual algorithms in the library.
family: Returns the family value from above
fitLibrary: A list with the fitted objects for each algorithm in SL.library on the full training data set.
varNames: A character vector with the names of the variables in X.
validRows: A list containing the row numbers for the V-fold cross-validation step.
method: A list with the method functions.
whichScreen: A logical matrix indicating which variables passed each screening algorithm.
control: The control list.
cvControl: The cvControl list.
errorsInCVLibrary: A logical vector indicating if any algorithms experienced an error within the CV step.
errorsInLibrary: A logical vector indicating if any algorithms experienced an error on the full data.

Details

recombineSL re-fits the super learner prediction algorithm using a new metalearning method. The weights for each algorithm in SL.library are re-estimated using the new metalearner, however the base learner fits are not regenerated, so this function saves a lot of computation time as opposed to using the SuperLearner function with a new method argument. The output is identical to the output from the SuperLearner function.

References

van der Laan, M. J., Polley, E. C. and Hubbard, A. E. (2008) Super Learner, Statistical Applications of Genetics and Molecular Biology, 6, article 25. http://www.bepress.com/sagmb/vol6/iss1/art25

Examples

Run this code

## Not run: 
# 
# # Binary outcome example adapted from SuperLearner examples
# 
# set.seed(1)
# N <- 200
# X <- matrix(rnorm(N*10), N, 10)
# X <- as.data.frame(X)
# Y <- rbinom(N, 1, plogis(.2*X[, 1] + .1*X[, 2] - .2*X[, 3] + 
#   .1*X[, 3]*X[, 4] - .2*abs(X[, 4])))
# 
# SL.library <- c("SL.glmnet", "SL.glm", "SL.knn", "SL.gam", "SL.mean")
# 
# # least squares loss function
# set.seed(1) # for reproducibility
# fit_nnls <- SuperLearner(Y = Y, X = X, SL.library = SL.library, 
#   verbose = TRUE, method = "method.NNLS", family = binomial())
# fit_nnls
# #                    Risk       Coef
# # SL.glmnet_All 0.2439433 0.01293059
# # SL.glm_All    0.2461245 0.08408060
# # SL.knn_All    0.2604000 0.09600353
# # SL.gam_All    0.2471651 0.40761918
# # SL.mean_All   0.2486049 0.39936611
# 
# 
# # negative log binomial likelihood loss function
# fit_nnloglik <- recombineSL(fit_nnls, Y = Y, method = "method.NNloglik")
# fit_nnloglik
# #                    Risk      Coef
# # SL.glmnet_All 0.6815911 0.1577228
# # SL.glm_All    0.6918926 0.0000000
# # SL.knn_All          Inf 0.0000000
# # SL.gam_All    0.6935383 0.6292881
# # SL.mean_All   0.6904050 0.2129891
# 
# # If we use the same seed as the original `fit_nnls`, then
# # the recombineSL and SuperLearner results will be identical
# # however, the recombineSL version will be much faster since
# # it doesn't have to re-fit all the base learners.
# set.seed(1)
# fit_nnloglik2 <- SuperLearner(Y = Y, X = X, SL.library = SL.library,
#   verbose = TRUE, method = "method.NNloglik", family = binomial())
# fit_nnloglik2
# #                    Risk      Coef
# # SL.glmnet_All 0.6815911 0.1577228
# # SL.glm_All    0.6918926 0.0000000
# # SL.knn_All          Inf 0.0000000
# # SL.gam_All    0.6935383 0.6292881
# # SL.mean_All   0.6904050 0.2129891
# 
# ## End(Not run)

Run the code above in your browser using DataLab