fitDPMM: Fit truncated joint DPMM to multiple time series
In lvhoskovec/psbpHMM: Covariate-dependent iHMMs for multiple time series

Fit truncated joint DPMM to multiple time series

fitDPMM(
  niter,
  nburn,
  y,
  ycomplete = NULL,
  priors = NULL,
  K.start = NULL,
  z.true = NULL,
  lod = NULL,
  mu.true = NULL,
  missing = FALSE,
  tau2 = NULL,
  a.tune = NULL,
  b.tune = NULL,
  resK = FALSE,
  eta.star = NULL,
  len.imp = NULL,
  holdout = NULL
)

`niter`	number of total iterations
`nburn`	number of burn-in iterations
`y`	list of time series data for each time series
`ycomplete`	complete data, if available, for evaluating imputations
`priors`	list of priors
`K.start`	maximum allowable number of states
`z.true`	list of true hidden states, if known
`lod`	list of lower limits of detection for p exposures for each time series
`mu.true`	matrix of true exposure means for each true state, if known
`missing`	logical; if TRUE then the data set contains missing data, default is FALSE
`tau2`	variance tuning parameter for normal proposal in MH update of lower triangular elements in decomposition of Sigma
`a.tune`	shape tuning parameter for inverse gamma proposal in MH update of diagonal elements in decomposition of Sigma
`b.tune`	rate tuning parameter for inverse gamma proposal in MH update of diagonal elements in decomposition of Sigma
`resK`	logical; if TRUE a resolvent kernel is used in MH update for lower triangular elements in decomposition of Sigma
`eta.star`	resolvent kernel parameter, must be a real value greater than 1. In the resolvent kernel we take a random draw from the geometric distribution with mean (1-p)/p, eta.star = 1/p.
`len.imp`	number of imputations to save. Imputations will be taken at equally spaced iterations between nburn and niter.
`holdout`	list of indicators of missing type in holdout data set, 0 = observed, 1 = MAR, 2 = below LOD, for imputation validation purposes

an object of type "dpmm"

a list with components

z.save: list of estimated hidden states for each time series at each iteration
K.save: list of estimated number of hidden states for each time series at each iteration
ymar: matrix of imputed values for MAR data, number of rows equal to len.imp
ylod: matrix of imputed values for data below LOD, number of rows equal to len.imp
hamming: posterior hamming distance between true and estimated states, if z.true is given
mu.mse: mean squared error for estimated state-specific means, if mu.true is given
mar.mse: mean squared error of MAR imputations, if ycomplete is given
lod.mse: mean squared error of imputations below LOD, if ycomplete is given
mismat: list, each element is a matrix indicating types of missing data for each time series, 0 = complete, 1 = MAR, 2 = below LOD
ycomplete: complete data
MH.arate: average MH acceptance rate for lower triangular elements
MH.lamrate: average MH acceptance rate for diagonal elements