compute_KLdiv.R

# This is a short script that shows how to compute the KL divergence using the kernel density estimator
# The key point is to use linear interpolation to evaluate the density at the same points

# 1. Generate a sample of standard normal variates
x <- rnorm(100)

# 2. Compute kernel density estimator
dens_obs <- density(x)

# 3. Use linear interpolation to evaluate over a grid
grid <- seq(-2, 2, length.out = 100)
dens_obs_linInt <- approxfun(dens_obs)(grid)

# 4. Compute KL divergence using flexmix package
mat <- cbind(obs = dens_obs_linInt,
             theo = dnorm(grid))
flexmix::KLdiv(mat)

Credit goes to this Stats SE answer: https://stats.stackexchange.com/questions/78711/how-to-find-estimate-probability-density-function-from-density-function-in-r