In ZijunGao/LinCDE: Conditional Density Estimation via Lindsey's Method and Boosting

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>",
  fig.path = "man/figures/README-",
  out.width = "100%"
)

Overview

Conditional density estimation is a fundamental problem in statistics, with various scientific and practical applications such as genetics and economics. We propose a conditional density estimator based on tree boosting and Lindsey's method (LinCDE). LinCDE admits flexible modeling of the density family and captures distributional characteristics like modality and shape. In particular, LinCDE always produces smooth and non-negative density estimates. Furthermore, in the presence of nuisance features, LinCDE identifies the influential covariates to the response distribution.

Installation

You can install the development version of LinCDE from GitHub with:

# install.packages("devtools")
# devtools::install_github("ZijunGao/LinCDE", build_vignettes = TRUE)

Example

Below is a basic example of using LinCDE for conditional mean estimation. We generate data $(y, X)$ where the conditional distributions of $y$ are Gaussian with different means and variances. The conditional mean depends on $X_1$ and $X_2$, and the conditional variance depends on $X_2$.

library(LinCDE)
# true conditional density function; LGD stands for locally Gaussian design
density.LGD = function(X, y = NULL){
  if(is.null(dim(X))){X = matrix(X, nrow = 1)}
  if(!is.null(y)){
    dens = dnorm(y, mean = (0.5 * X[, 1] + X[, 1] * X[, 2]), sd = (0.5 + 0.25 * X[, 2]))
  }else{
    y = 0.5 * X[, 1] + X[, 1] * X[, 2] + rnorm(dim(X)[1], 0, 1) * (0.5 + 0.25 * X[ ,2])
  }
}

# training data
set.seed(100)
n = 1000; d = 20
X = matrix(runif(n * d, -1, 1), ncol = d); colnames(X) = paste("X", seq(1,d), sep = "")
y.LGD = density.LGD(X)

We fit a LinCDE model with default hyper-parameters. Details of hyper-parameter tuning are discussed in the vignette.

# fit a LinCDE boost model
model.example = LinCDE.boost(X = X, y = y.LGD, verbose = F)

We plot the estimated conditional densities from LinCDE against the true conditional densities at $9$ landmarks. LinCDE's fit is reasonably close to the truth.

# landmarks for visualization
XProfile = matrix(0, nrow = 3^2, ncol = d); colnames(XProfile) = paste("X", seq(1,d), sep = "")
X1 = X2 = c(-0.6,0,0.6); XProfile[, c(1,2)] = as.matrix(expand.grid(X1, X2))

# visualize LinCDE boost model's predictions at the landmarks
densityPlot(X = XProfile, trueDensity = density.LGD, model = model.example)

For more examples, please see the vignette :)

ZijunGao/LinCDE documentation built on Jan. 2, 2023, 11:14 p.m.