Nonparametric estimation for probability mass function with Disake: an R package for discrete associated kernel estimators

Abstract

International audience Kernel smoothing is one of the most widely used nonparametric data smoothing techniques. We introduce a new R package, Disake, for computing discrete associated kernel estimators for probability mass function. When working with a kernel estimator, two choices must be made: the kernel function and the smoothing parameter. The Disake package focuses on discrete associated kernels and also on cross-validation and local Bayesian techniques to select the appropriate bandwidth. Applications on simulated data and real data show that the binomial kernel is appropriate for small or moderate count data while the empirical estimator or the discrete triangular kernel is indicated for large samples.

Kernel density estimation of allele frequency including undetected alleles

“…KDE is a method to estimate the probability density function by assuming the potential probability for the range around the detected data (samples) and then summing the probabilities. KDE is also used to estimate the probability mass function ( e.g ., Wansouwé, Kokonendji & Kolyang, 2015 ). We found an analogy between ordinal KDE and the estimation of allele frequency, including that of undetected alleles: First, both the allele frequency and probability density function sum up to one, and their portion is always non-negative.…”