Ordered choice models now include an intercept by default.
generate_choice_parameters() now draws missing coefficients from a standard
normal distribution instead of a normal distribution with variance 10.
Added support for uncorrelated normal and log-normal random effects.
Added latent_class_effects to choice_formula(), which names the effects
that differ between latent classes, and the column latent_class to
choice_effects().
A named beta is matched to the effects by name instead of by position,
and generate_choice_parameters() completes a named beta that fixes only
some of the effects.
Probit probabilities now come from oeli::pmvnorm().
Specialized choiceprob_*() functions are now internal; use
compute_choice_probabilities() as the public interface.
Improved input validations. Internal validation failures now use descriptive
cli messages.
Added partial rankings, individual choice sets, joint panel outcomes, log-normal random effects, and latent class estimation.
Probability calculations now use Rcpp.
Added the choice data sets water_conservation_choice, wind_power_choice,
and lichess_berserk_choice.
long_to_wide() now accepts a factor column of alternatives.
J in choice_alternatives() defaults to the number of supplied choice
alternatives.
New function train_test(), which splits choice data by deciders or by choice
occasions into a train and a test subset.
Removed unused package dependencies.