Efron (2016, 2019) penalized logspline density estimator for Gaussian mixture model g-modeling. Returns an object of class GLmix to facilitate prediction compatible with Kiefer-Wolfowitz GLmix estimation. In particular percentile confidence intervals can be constructed based on posterior quantiles. Assumes homoscedastic standard Gaussian noise, for the moment.
BDGLmix(y, T = 300, sigma = 1, df = 5, c0 = 0.1)An object of class GLmix, density with components:
points of evaluation on the domain of the density
estimated function values at these points of the mixing density
returns a sigma = 1 for compatibility with GLmix
Data: Sample Observations
Undata: Grid Values defaults equal spacing of with T bins, when T is a scalar
scale parameter of the Gaussian noise, may take vector value of length(y)
degrees of freedom of the natural spline basis
penalty parameter for the Euclidean norm penalty.
Adapted from a similar implementation in the R package deconvolveR of Narasimhan and Efron.
Efron, B. (2016) Empirical Bayes deconvolution estimates, Biometrika, 103, 1–20, Efron, B. (2019) Bayes, Oracle Bayes and Empirical Bayes, Statistical Science, 34, 177-201.