mnl_mixture_coord_exch: Coordinate exchange algorithm for a Multinomial Logit Scheffé...
In mariobecerra/opdesmixr: Optimal experimental designs for mixture models

Description Usage Arguments Details Value

Performs the coordinate exchange algorithm for a Multinomial Logit Scheffé model as described in Bayesian D-optimal choice designs for mixtures by Ruseckaite, Goos & Fok (2017).

mnl_mixture_coord_exch(
  q = NULL,
  J = NULL,
  S = NULL,
  n_random_starts = 100,
  X = NULL,
  beta,
  transform_beta = T,
  order = 3,
  opt_method = "B",
  max_it = 10,
  tol = 1e-04,
  n_cox_points = NULL,
  plot_designs = F,
  verbose = 1,
  opt_crit = 0,
  seed = NULL,
  n_cores = 1,
  save_all_designs = F
)

`q`	number of mixture ingredient proportions.
`J`	number of alternatives within a choice set.
`S`	number of choice sets.
`n_random_starts`	number or random starts. Defaults to 100.
`X`	If an initial design is to be supplied, then it must be a 3-dimensional array of size `(q, J, S)`, with q, J, and S defined above.
`beta`	Prior parameters. For a locally optimal design, it should be a numeric vector of length m = (q^3 + 5*q)/6. For a pseudo-Bayesian design, it must be a matrix with prior simulations of size (nxm) where m is previously defined and m is the number of prior draws, i.e., there is a prior draw per row.
`transform_beta`	boolean parameter. Should the `beta` vector/matrix be transformed by subtracting the q-th element?
`order`	integer corresponding to a Scheffé model order (1, 2, 3).
`opt_method`	Optimization method in each step of the coordinate exchange algorithm. It can be "B" (Brent's algorithm) or "D" (discretization of Cox direction)
`max_it`	integer for maximum number of iterations that the coordinate exchange algorithm will do
`tol`	A positive error tolerance in Brent's method.
`n_cox_points`	number of points to use in the discretization of Cox direction. Ignored if opt_method is Brent.
`plot_designs`	boolean. If TRUE, shows a plot of the initial and the final design. Only works if q is 3 or 4.
`verbose`	level of verbosity. See below for details.
`opt_crit`	optimality criterion: D-optimality ("D" or 0) or I-optimality ("I" or 1).
`seed`	Seed for reproducibility.
`n_cores`	Number of cores for parallel processing.
`save_all_designs`	Whether the function should return a list with all the designs created at random or only the best.

Verbosity levels: each level prints the previous plus additional things:

Print the efficiency value in each iteration and a final summary
Print the values of k, s, i, and efficiency value in each subiteration
Print the resulting X after each iteration, i.e., after each complete pass on the data
Print efficiency value for each point in the Cox direction discretization
Print the resulting X and information matrix after each subiteration
Print the resulting X or each point in the Cox direction discretization

The function returns a list with 11 elements:

X_orig: The original design. A 3-dimensional array of size (q, J, S).
X: The optimized design. A 3-dimensional array of size (q, J, S).
beta: The original beta vector or matrix.
opt_crit_value_orig: efficiency of the original design.
opt_crit_value: efficiency of the optimized design.
n_iter: Number of iterations performed.
efficiency_value_per_iteration: Efficiency value in each iteration of the algorithm.
opt_crit: The optimality criterion used.
q: Number of mixture ingredients.
seed: seed used to generate the final design. If a design was used as input by the user, this will be NA.