In gbasulto/kerdec: Kernel deconvolution methods

rm(list=ls()) ### To clear namespace
library(knitr)

# opts_chunk$set(fig.width=12, fig.height=8, fig.path='Figs/',
#                echo=TRUE, warning=FALSE, message=FALSE)
opts_chunk$set(fig.width= 7, fig.height= 5,
               echo=TRUE, warning=FALSE, message=FALSE)

Bandwidth Selection in KDDE

Bandwidth selection is a crucial aspect while obtaining kernel deconvolution density estimates (KDDEs). It is not straightforward to adapt existing methods in usual kernel density estimators (KDEs) to their counterpart for samples contaminated with error, that is, KDDEs.

The kerdec packages allows the user to either provide the bandwidth, or select some common methods that exist for KDDEs. In the bivariate setting, the options are more limited.

Bandwidth Given

library(kerdec)                         # Load packages
library(ggplot2)                         
library(dplyr)
library(purrr)

set.seed(666)                           # Set seed


## Signal parameters 
shape <- 5                              # X distr. shape par.
rate <- 5                               # X distr. rate par.

## Settings and samples for all the cases
n <- 150                                # Sample size
sd_error <- 0.2                         # std. error of error distr.
X <- rgamma(n, shape, rate)             # Uncontaminated sample
Y <- X + rlaplace(n, sd = sd_error)     # Contaminated sample


## Estimate density
get_KDDE_and_return_df <- function(h)
{
  kdde <- kerdec_dens(smp = Y, error_scale_par = sd_error, h = h)
  out <- data_frame(x = kdde$x_eval, value = kdde$f_vals, bandwidth = as.character(h))
  return (out)
}

h <- c(0.1, 0.2, 0.08)
out <- map_df(h, get_KDDE_and_return_df)

ggplot(out, aes(x, value, color = bandwidth)) + geom_line(aes(linetype = bandwidth))