fix NA handling in build_mm

DESCRIPTION

@@ -1,7 +1,7 @@
 Package: seqHMM
 Title: Mixture Hidden Markov Models for Social Sequence Data and Other Multivariate, Multichannel Categorical Time Series
 Version: 1.2.6
-Date: 2023-06-14
+Date: 2023-06-7
 Authors@R: 
  c(person(given = "Jouni",
  family = "Helske",

NEWS.md

@@ -1,9 +1,11 @@
-seqHMM 1.2.6 (Release date: )
+seqHMM 1.2.6 (Release date: 2023-06-07)
 ==============
  * Armadillo solver now fails directly instead of trying approximate solution 
  in case of numerical issues when estimating the regression coefficients.
+ * `build_mm` now correctly runs EM algorithm only in case of real missing 
+ observations in data, i.e. it correctly handles void values.
 
-seqHMM 1.2.5 (Release date: 2023-05-30)
+seqHMM 1.2.5 (Release date: 2023-06-13)
 ==============
  * New methods `state_names` and `cluster_names` for getting and setting state 
  and cluster names of the model objects.

R/build_mm.R

@@ -77,20 +77,15 @@ build_mm <- function(observations) {
  state_names <- alphabet(observations)
  n_states <- length(state_names)
 
- nobs <- sum(!(observations == attr(observations, "nr") |
- observations == attr(observations, "void") |
- is.na(observations)))
-
-
- if (nobs < prod(dim(observations))) {
+ if (any(observations == attr(observations, "nr"))) {
  model <- build_hmm(
  observations,
  transition_probs = matrix(1 / n_states, n_states, n_states),
  emission_probs = diag(n_states),
  initial_probs = rep(1 / n_states, n_states),
  state_names = state_names)
  model <- fit_model(model)$model
- message("Sequences contain missing/void values, initial and transition probabilities estimated via EM. ")
+ message("Sequences contain missing values, initial and transition probabilities estimated via EM. ")
  } else {
  first_timepoint <- suppressMessages(seqdef(observations[observations[, 1] %in% state_names, 1], alphabet = state_names))
  initial_probs <- TraMineR::seqstatf(first_timepoint)[, 2] / 100