This is an R package implementing DDR, an algorithm that esimates conditional density functions by transforming the response variable of regression into a set of asymptotically Dirac delta functions using kernel density functions. This allows the user to convert a non-linear regressor into a conditional density estimator. We use kernel ridge regression as the underlying regressor in this implementation.

The academic article describing DDR in detail can be found here. Please cite the article if you use any of the code in this repository.

Please install the FNN, pracma, doParallel, Rfast and foreach packages on CRAN. Then:

library(devtools)

install_github("ericstrobl/DDR")

library(DDR)

numCores <- detectCores()-1; registerDoParallel(numCores) # set up parallel computing

X = matrix(rnorm(400),200,2); y = X[,1]+rnorm(200); X_te = matrix(rnorm(20),10,2) # generate data

cd_est = DDR(X,y,X_te) # run DDR

plot(cd_est$y,cd_est$dens[1,],type="l") # plot the conditional density estimate of the first test sample

lines(cd_est$y,dnorm(cd_est$y,X_te[1,1]),col="red") # plot ground truth in red

closeAllConnections() # close parallel computing

Recommended if you know that the response variable Y is bounded on an interval [lb,ub]. Default is lb=-Inf and ub=Inf.

X = matrix(rnorm(400),200,2); y = rbeta(200,0.5,0.5); X_te = matrix(rnorm(20),10,2) # generate data with response from a beta distribution (alpha=0.5, beta=0.5)

cd_est = DDR(X,y,X_te,lb=0,ub=1) # run DDR

plot(cd_est$y,cd_est$dens[1,],type="l") # plot the conditional density estimate of the first test sample

lines(cd_est$y,dbeta(cd_est$y,0.5,0.5),col="red") # plot ground truth in red