The frozen mixqr engine contract (documentation only) — mixqr_engine

A mixqr_engine is a list of closures with these signatures. The generic driver mixqr_em() owns convergence, multi-start, labelling and bookkeeping; engines supply only the statistical updates.

Details

estep(state, X, y, tau, weights): returns an n x m responsibility matrix p whose rows sum to 1.
mstep_pi(p, weights): returns a length-m mixing-probability vector.
mstep_beta(p, X, y, tau, weights, beta_prev): returns a (ncol(X)) x m coefficient matrix (weighted quantile regression).
update_density(p, X, y, beta, tau, control): returns a length-m list of density objects, each a list with at least eval (a function), f0 (density at 0) and type.
loglik(state, X, y, tau): returns the log-likelihood or NA.
npar(state): returns the number of free parameters.

state is a list with elements pi, beta, density, tau, m, error_density.