uduwage/medley

A system for blending regression models in R

What is Medley?

Medley is an R package which implements the Caruana et al., 2004 algorithm for greedy stepwise ensemble selection for regression models.

The idea behind medley is to make the creation of "ensembles" or "blends" of models as simple as possible. Individual models can be produced by varying hyperparameters or input feature sets, as well as by changing the underlying model code.

Usage example

require(medley);
require(randomForest);
require(e1071);

# x and y are the training predictors and responses respectively
m <- create.medley(x, y, errfunc=rmse);

# add SVMs for a variety of gamma parameters
for (g in 1:10) {
  m <- add.medley(m, svm, list(gamma=1e-3 * g));
}

# add random forests with varying mtry parameter
for (mt in c(5, 10, 20, 50)) {
  m <- add.medley(m, randomForest, list(mtry=mt));
}

# use only the best 80% of individual models
m <- prune.medley(m, 0.8);

# predict using new predictor matrix newx
p <- predict(m, newx);