multinomial_em function

EM algorithm for multinomial data

Implement the EM algorithm for multivariate multinomial data given observed counts of complete and missing data ($Y_obs$ and $Y_mis$). Allows for specification of a Dirichlet conjugate prior.

multinomial_em(
  x_y,
  z_Os_y,
  enum_comp,
  n_obs,
  conj_prior = c("none", "data.dep", "flat.prior", "non.informative"),
  alpha = NULL,
  tol = 5e-07,
  max_iter = 10000,
  verbose = FALSE
)

Arguments

• x_y: A data.frame of observed counts for complete observations.
• z_Os_y: A data.frame of observed marginal-counts for incomplete observations.
• enum_comp: A data.frame specifying a vector of all possible observed patterns.
• n_obs: An integer specifying the number of observations in the original data.
• conj_prior: A string specifying the conjugate prior. One of c("none", "data.dep", "flat.prior", "non.informative").
• alpha: The vector of counts $\alpha$ for a $Dir(\alpha)$ prior. Must be specified if conj_prior is either c("data.dep", "flat.prior"). If flat.prior, specify as a scalar. If data.dep, specify as a vector with key matching enum_comp.
• tol: A scalar specifying the convergence criteria. Defaults to 5e-7
• max_iter: An integer specifying the maximum number of allowable iterations. Defaults to 10000.
• verbose: Logical. If TRUE, provide verbose output on each iteration.

Returns

An object of class mod_imputeMulti-class.

Examples

## Not run:

 data(tract2221)
 x_y <- multinomial_stats(tract2221[,1:4], output= "x_y")
 z_Os_y <- multinomial_stats(tract2221[,1:4], output= "z_Os_y")
 x_possible <- multinomial_stats(tract2221[,1:4], output= "possible.obs")

 imputeEM_mle <- multinomial_em(x_y, z_Os_y, x_possible, n_obs= nrow(tract2221),
                     conj_prior= "none", verbose= TRUE)
## End(Not run)

See Also

multinomial_data_aug, multinomial_impute