mgpsEM: Multi-Item Gamma-Poisson Shrinker (MGPS)

In sidiwang/hgzips: hgzips

Description

This mgpsEM function finds hyperparameter estimates by implementing the Expectation-Maximization (EM) algorithm.nlminb function is used to maximize the loglikelihood function.

Usage

mgpsEM(
  alpha1,
  beta1,
  alpha2,
  beta2,
  pi,
  N,
  E,
  weight,
  iteration,
  Loglik = FALSE,
  zeroes = FALSE,
  N_star = 1
)

Arguments

alpha1	initial shape parameter guess of the first gamma distribution
beta1	initial rate parameter guess of the first gamma distribution
alpha2	initial shape parameter guess of the second gamma distribution
beta2	initial rate parameter guess of the second gamma distribution
pi	initial mixing proportion guess of the two gamma distributions
N	vector of Nij values
E	vector of Eij values
weight	set weight = rep(1, length(N)) if N and E are not squashed data, or input the weight vector corresponding to the squashed Nij vector.
iteration	number of EM algorithm iterations to run
Loglik	whether to return the loglikelihood of each iteration or not (TRUE or FALSE)
zeroes	A logical scalar specifying if zero counts should be included.
N_star	the minimum Nij count size to be used for hyperparameter estimation. If zeroes are included in Nij vector, please set N_star = NULL

Value

a list including the following:

• theta_EM Estimate of hyperparameters for each EM iteration

• llh logliklihood for each EM iteration (optional)

References

DuMouchel W (1999). "Bayesian Data Mining in Large Frequency Tables, With an Application to the FDA Spontaneous Reporting System." The American Statistician, 53(3), 177-190.

See Also

openEBGM, nlminb

Examples

Nij = rnbinom(100, size = 2, prob = 0.3)
Eij = runif(100, 0, 2)
par_estimated = mgpsEM(2, 4, 0.2, 0.1, 0.33, Nij, Eij, rep(1, length(Nij)), iteration = 10, Loglik = TRUE, zeros = TRUE, N_star = 1)
par_estimated$theta_EM
par_estimated$Loglik

sidiwang/hgzips documentation built on Jan. 19, 2021, 4:09 p.m.