This repo contains an R implementation of the ADEBA density estimation method and a benchmark study comparing it to a number of other methods.
Original publication here. Notice that the link only access a preview of the paper, which is accepted for publication in Pattern Recognition, Volume 78, June 2018, Pages 133-143.
Latest release version:
R> install.packages("adeba")
Latest dev version (currently the same):
devtools::install_github("backlin/adeba/adeba")
Some extra plot functionality is provided in the support package
adebaExtra
(not on CRAN).
devtools::install_github("backlin/adeba/adebaExtra")
The ADEBA family of estimators has two features that differentiates it from traditional estimators (those included in the base distributions of R and Matlab). These are demonstrated below.
ADEBA has two hyperparameters, one that sets the overall bandwidth and one that adjusts it to the local data density of each kernel. This allows it to:
Capture sharp modes:
pic here plz
Remove spurious modes in the tails:
ADEBA extends naturally to multivariate densities.