knitr::opts_chunk$set( collapse = TRUE, comment = "#>", fig.path = "README-" )
r2weight
Machine learning-based summary of association with multivariate outcomes
Authors: David Benkeser
This package provides a method for summarizing the strength of association between a set of variables and a multivariate outcome. In particular, cross-validation is combined with stacked regression (aka super learning) to estimate the convex combination of a multivariate outcome that maximizes cross-validated R-squared of a super learner-based prediction. The method is particularly well suited for situations with high-dimensional covariates and/or complex relationships between covariates and outcomes.
You can install a stable release of r2weight
from GitHub via
devtools
with:
devtools::install_github("benkeser/r2weight")
In the future, the package will be available from CRAN via
install.packages("r2weight")
The basic workflow of the package is to call the function optWeight
, which estimates the optimal weights for the combined outcome. The function r2_optWeight
is then be called using the optWeight
object to obtain a cross-validated estimate of the R-squared for predicting the optimally combined outcome. This measure provides a summary of the strength of association between covariates and outcome.
Here we illustrate the method using simulated data.
# sample size n <- 500 # set the seed set.seed(12345) # simulate nine covariate predictors x1 <- runif(n,0,4) x2 <- runif(n,0,4) x3 <- runif(n,0,4) x4 <- rbinom(n,1,0.75) x5 <- rbinom(n,1,0.25) x6 <- rbinom(n,1,0.5) x7 <- runif(n,0,4) x8 <- runif(n,0,4) x9 <- runif(n,0,4) # put all predictors in single data.frame X <- data.frame(x1=x1, x2=x2, x3=x3, x4=x4, x5=x5, x6=x6, x7=x7, x8=x8, x9=x9) # simulate three outcomes y1 <- x1 + 2*x2 + 4*x3 + x4 + 2*x5 + 4*x6 + 2*x7 + rnorm(n, 0, 5) y2 <- x1 + 2*x2 + 4*x3 + x4 + 2*x5 + 4*x6 + 2*x8 + rnorm(n, 0, 5) y3 <- x1 + 2*x2 + 4*x3 + x4 + 2*x5 + 4*x6 + 2*x9 + rnorm(n, 0, 5) # put all outcomes in single data.frame Y <- data.frame(y1 = y1, y2 = y2, y3 = y3) # call optWeight using simple Super Learner library out1 <- optWeight(Y = Y, X = X, SL.library = c("SL.mean","SL.glm","SL.step")) # print the object out1
The estimated optimal weights are shown along with the estimated cross validated R-squared for predicting each outcome using the Super Learner. If desired, there is an S3-method for predicting the combined outcome on new data.
# generate new data newX <- data.frame(x1=1, x2=1, x3=1, x4=1, x5=1, x6=1, x7=1, x8=1, x9=1) # get prediction of combined outcome predict(out1, newdata = newX)
## [,1] ## [1,] 15.80969
Next, we call r2_optWeight
to estimate the cross-validated R-squared for predicting the combined outcome with Super Learner.
# cross-validated R-squared # set verbose = TRUE to see a progress bar r2.out1 <- r2_optWeight(out1, Y = Y, X = X) # print the output r2.out1
The R-squared for each individual outcome is shown, as well as for the combined outcome. In this example, the true R-squared for individual outcomes is about 0.6 and for the combined outcome about 0.8.
A measure of variable importance for a particular variable can be defined as the difference in R-squared for the combined outcome when including and excluding that variable. These measures can be estimated using the r2_diff
function.
# measure importance of x9 # call optWeight excluding x9 from X out2 <- optWeight(Y = Y, X = X[,1:8], SL.library = c("SL.glm","SL.mean")) # print output out2 # compare to full fit to get importance for each individual outcome outDiff <- r2_diff(out1, out2) # difference in R-squared for first outcome # with confidence interval and p-value for two-sided test that # the difference equals 0 outDiff$y1$diff # for the third outcome outDiff$y3$diff # for combined outcome r2.out2 <- r2_optWeight(out2, Y = Y, X = X[,1:8]) # print output, notice change in weights r2.out2 # compare to full fit to get importance for combined outcome outDiff.r2 <- r2_diff(r2.out1, r2.out2) # print output outDiff.r2
© 2016-2017 David C. Benkeser
The contents of this repository are distributed under the MIT license. See below for details:
The MIT License (MIT) Copyright (c) 2016-2017 David C. Benkeser Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions: The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.