Maximum likelihood for SIRD-GA from daily cases and/or deaths

Nelder-Mead maximisation of sird_ga_loglik_inc() on an unconstrained scale (log rates, logit probabilities and fractions), from one or several starting points; the best optimum is polished by a final BFGS run. Parameters not listed in free are held at their values in fixed.

Usage

sird_ga_fit_inc(
  cases = NULL,
  deaths = NULL,
  N,
  use = c("both", "cases", "deaths"),
  starts,
  free = c("lambda", "gamma", "alpha", "i0", "p", "tau2", "tau2_d"),
  fixed = list(s0 = NULL, r0 = 0, q = 1, p = 1, tau2_d = 0),
  p_max = 0.999,
  Delta = 1,
  nsub = 10L,
  maxit = 4000,
  hessian = TRUE
)

Arguments

cases, deaths, N, use, Delta, nsub: See sird_ga_loglik_inc().
starts: A list of named lists of starting values (one per start).
free: Character vector of parameters to estimate, a subset of c("lambda", "gamma", "alpha", "i0", "s0", "p", "tau2", "q", "tau2_d").
fixed: Named list with values for the remaining parameters (r0 is always derived as 1 - s0 - i0 when s0 is free, otherwise taken from fixed).
p_max: Upper bound for p.
maxit: Iteration limit for each Nelder-Mead run.
hessian: Logical; return the numerical Hessian (unconstrained scale) at the optimum.

Value

A list with theta, loglik, starts (data frame of the optimum reached from each start), hessian, free and the transformation functions.