Learning with Subset Stacking (R version)
LESS is a supervised learning algorithm that is based on training many local estimators on subsets of a given dataset, and then passing their predictions to a global estimator. You can find the details about LESS in our manuscript.
install.packages("less")
library(less)
Here is how you can use LESS:
install.packages("caret")
install.packages("MLmetrics")
library(caret)
library(MLmetrics)
### CLASSIFICATION ###
# Load data
data(iris)
# Train and test split
split_list <- train_test_split(iris, test_size = 0.3)
X_train <- split_list[[1]]
X_test <- split_list[[2]]
y_train <- split_list[[3]]
y_test <- split_list[[4]]
# LESS fit() & predict()
lessclassifier <- LESSClassifier$new(random_state = 2022)
lessclassifier$fit(X_train, y_train)
preds <- lessclassifier$predict(X_test)
table <- caret::confusionMatrix(data=factor(preds), reference = factor(y_test))
cat("Test accuracy of LESS: ", table$overall[["Accuracy"]])
### REGRESSION ###
# Load data
data(abalone)
# Train and test split
split_list <- train_test_split(abalone, test_size = 0.3)
X_train <- split_list[[1]]
X_test <- split_list[[2]]
y_train <- split_list[[3]]
y_test <- split_list[[4]]
# LESS fit() & predict()
less_model <- LESSRegressor$new(random_state = 2022)
less_model$fit(X_train, y_train)
y_pred <- less_model$predict(X_test)
mse <- MLmetrics::MSE(y_pred, y_test)
cat("Test error of LESS: ", mse)
Our two-part tutorial on Colab aims at getting you familiar with LESS regression.
Default implementation of LESS uses Euclidean distances with radial basis function. Therefore, it is a good idea to scale the input data before fitting. This can be done by setting the parameter scaling in LESSRegressor or LESSClassifier to TRUE (this is the default value)
Python implementation of LESS is available in another repository.
Citation for our software can be reached by:
citation("less")
or
To cite package ‘less’ in publications use:
Birbil I, Cavdar B (2022). _less: Learning with Subset Stacking_. R package version 0.1.0, <https://CRAN.R-project.org/package=less>.
A BibTeX entry for LaTeX users is
@Manual{,
title = {less: Learning with Subset Stacking},
author = {Ilker Birbil and Burhan Ozer Cavdar},
year = {2022},
note = {R package version 0.1.0},
url = {https://CRAN.R-project.org/package=less},
}