mape_negbinomial: mape_negbinomial

In alexandercoulter/DHBayes: Bayesian Hierarchical Sampler for Discrete (Binomial, Negative-Binomial, Poisson) Data

Description

mape_negbinomial

Usage

mape_negbinomial(
  df,
  r_prior,
  p_prior,
  M_prior,
  eta = 0.1,
  tol = 1e-04,
  maxIter = 10000,
  method = c("newton", "gdescent")
)

Arguments

df	data.frame object, containing at least column named 'x' containing non-negative integer values.
r_prior	positive numeric; recommended choose as if there was observed "prior sample" Y of size 'M_prior', where Y ~ nbinom(r_prior, p_prior).
p_prior	numeric strictly greater than 0 and less than 1; recommended choose as if there was observed "prior sample" Y of size 'M_prior', where Y ~ nbinom(r_prior, p_prior).
M_prior	positive numeric; recommended choose as if there was observed "prior sample" Y of size 'M_prior', where Y ~ nbinom(r_prior, p_prior).
eta	positive numeric dampening parameter for Newton's method, gradient descent algorithm.
tol	non-negative numeric tolerance parameter for exiting optimization algorithm.
maxIter	positive integer setting maximum number of iterations for optimization algorithm.
method	string controlling optimization method; default 'newton'.

Value

List object containing maximum a-posteriori estimates (MAPEs) of negative binomial distribution parameters r and p, assuming df$x ~ nbinom(r, p), and (r, p) ~ MVN(mu, Sigma). (Please see DHBayes_Derivations.pdf on GitHub for how mu, Sigma relate to 'r_prior', 'p_prior', and 'M_prior' inputs.)

Examples

# Generate example data:
set.seed(31)
r = 4
p = 0.3

# Number of experiments, i.e. rows in df:
numexps = 10

# Filling df with pseudo data; note the requisite column 'x':
df = data.frame('x' = rnbinom(numexps, r, p))

# Generating maximum a posteriori estimate (MAPE) solution for r and p:
rp_fit = mape_negbinomial(df = df, r_prior = 2, p_prior = 0.5, M_prior = 1)

# Compare fitted values to known values:
cbind(c(r, p), c(rp_fit$r, rp_fit$p))

alexandercoulter/DHBayes documentation built on Dec. 19, 2021, 12:29 a.m.