Each procedure’s probability mass function (PMF) and cumulative distribution function (CDF) was implemented in C++ using the Rcpp
package. By means of Rcpp::interface
, these functions are exported to both the package’s R namespace and C++ headers. That way, the following functions can then be used by other packages that use Rcpp
:
/*** Ordinary Poisson Binomial Distribution ***/
/*** Exact Procedures ***/
// Direct Convolution (DC)
NumericVector dpb_conv(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_conv(IntegerVector obs, NumericVector probs); // CDF
// Divide & Conquer FFT Tree Convolution (DC-FFT)
NumericVector dpb_dc(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_dc(IntegerVector obs, NumericVector probs); // CDF
// Discrete Fourier Transformation of the Characteristic Function (DFT-CF)
NumericVector dpb_dftcf(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_dftcf(IntegerVector obs, NumericVector probs); // CDF
// Recursive Formula (RF)
NumericVector dpb_rf(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_rf(IntegerVector obs, NumericVector probs); // CDF
/*** Approximations ***/
// Arithmetic Mean Binomial Approximation (AMBA)
NumericVector dpb_mean(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_mean(IntegerVector obs, NumericVector probs); // CDF
// Geometric Mean Binomial Approximations (GMBA)
NumericVector dpb_gmba(IntegerVector obs, NumericVector probs, bool anti); // PMF
NumericVector ppb_gmba(IntegerVector obs, NumericVector probs, bool anti); // CDF
// Poisson Approximation (PA)
NumericVector dpb_pa(IntegerVector obs, NumericVector probs); // PMF
NumericVector ppb_pa(IntegerVector obs, NumericVector probs); // CDF
// Normal Approximations (NA, RNA)
NumericVector dpb_na(IntegerVector obs, NumericVector probs, bool refined); // PMF
NumericVector ppb_na(IntegerVector obs, NumericVector probs, bool refined); // CDF
/*** Generalized Poisson Binomial Distribution ***/
/*** Exact Procedures ***/
// Generalized Direct Convolution (G-DC)
NumericVector dgpb_conv(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q); // PMF
NumericVector pgpb_conv(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q); // CDF
// Generalized Discrete Fourier Transformation of the Characteristic Function (G-DFT-CF)
NumericVector dgpb_dftcf(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q); // PMF
NumericVector pgpb_dftcf(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q); // CDF
/*** Approximations ***/
// Generalized Normal Approximations (G-NA, G-RNA)
NumericVector dgpb_na(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q, bool refined); // PMF
NumericVector pgpb_na(IntegerVector obs, NumericVector probs,
NumericVector val_p, NumericVector val_q, bool refined); // CDF
There are only a few simple steps to follow:
Rcpp
and PoissonBinomial
packages to the Imports
and LinkingTo
fields of the DESCRIPTION
file.#include <PoissonBinomial.h>
to source (.cpp
) and/or header (.h
, .hpp
) files in which these functions are to be used.using namespace PoissonBinomial;
. Without it, the use of functions of this package must be fully qualified with PoissonBinomial::
, e.g. PoissonBinomial::dpb_dc
instead of dpb_dc
For better performance, the PMFs and CDFs do not check any of their parameters for plausibility! This must be done by the user by means of R or C/C++ functions. It must be made sure that
obs
vectors are valid, i.e. in appropriate ranges,probs
vector are in \((0, 1)\) anddpb_gmba
, ppb_gmba
, dpb_na
, ppb_na
, dgpb_na
and pgpb_na
: the probabilities in the probs
vector must not contain zeros or ones.Furthermore, the CDFs only compute lower-tail cumulative probabilities, i.e. \(P(X \leq k)\). If upper-tail probabilities \(P(X > k)\) are needed, they must be computed “manually”. The same applies for logarithmic probabilities, which can be calculated using Rcpp::log
for instance.