The function ICA.control
returns a list of ICA control parameters.
ICA.control(
ncount = 40,
nimp = ncount/10,
assim_coeff = 4,
revol_rate = 0.3,
damp = 0.99,
uniting_threshold = 0.02,
equal_weight = FALSE,
sym = FALSE,
sym_point = NULL,
stop_rule = c("maxiter", "equivalence"),
stoptol = 0.99,
checkfreq = 0,
plot_cost = TRUE,
plot_sens = TRUE,
plot_3d = c("lattice", "rgl"),
trace = TRUE,
rseed = NULL
)
Number of countries. Defaults to 40
.
Number of imperialists. Defaults to 10 percent of ncount
.
Assimilation coefficient. Defaults to 4
.
Revolution rate. Defaults to 0.3
.
Damp ratio for revolution rate. revol_rate
is decreased in every iteration by a factor of damp
(revol_rate * damp
). Defaults to 0.99
.
If the distance between two imperialists is less than the product of the uniting threshold by the largest distance in the search space, ICA unites the empires. Defaults to 0.02
.
Should the weights of design points assumed to be equal? Defaults to FALSE
. If TRUE
, it reduces the dimension of the search space and produces a design that gives equal weight to all of its support points.
Should the design points be symmetric around sym_point
? Defaults to FALSE
. When TRUE
, sym_point
must be given.
If sym = TRUE
, the design points will be symmetric around sym_point
. See 'Details'.
Either 'maxiter'
or 'equivalence'
.
Denotes the type of stopping rule. See 'Details'. Defaults to 'maxiter'
.
If stop_rule = 'equivalence'
, algorithm stops when ELB is larger than stoptol
. Defaults to 0.99
.
The algorithm verifies the general equivalence theorem in
every checkfreq
iterations.
When checkfreq = 0
, no verification will be done. When checkfreq = Inf
, only the output design will be verified.
Defaults to 0
.
Plot the iterations (evolution) of algorithm? Defaults to TRUE
.
Plot the sensitivity (derivative) function at every checkfreq
. Defaults to TRUE
.
Character. Which package should be used to plot the sensitivity plot for models with two explanatory variables?
Print the information in every iteration? Defaults to TRUE
.
Random seed. Defaults to NULL
.
A list of ICA control parameters.
If stop_rule = 'maxiter'
, the algorithm iterates until maximum number of iterations.
If stope_rule = 'equivalence'
, the algorithm stops when either ELB is greater than stoptol
or it reaches maxiter
.
In this case, you must specify the check frequency by checkfreq
.
Note that checking equivalence theorem is a very time consuming process,
especially for Bayesian and minimax design problems.
We advise using this option only for locally, multiple objective and robust optimal designs.
What to follows shows how sym_point
and sym
may be useful?
Assume the 2PL model of the form \( P(Y=1) = \frac{1}{1+exp(-b(x - a))}\) and
let the parameters \(a\) and \(b\)
belong to
\([a_L, a_U]\) and \([b_L, b_U]\), respectively.
It can be shown that the optimal design for this model
is symmetric around \(a_M = \frac{a_L + a_U}{2}\).
For this model, to find accurate symmetric designs, one can set sym = TRUE
and
provide the value of the \(a_M\) via sym_point
.
In this case, the output design will be symmetric around the point sym_point
.
The length of sym_point
must be equal to the number of model predictors, here, is equal to 1
.
# NOT RUN {
ICA.control(ncount = 100)
# }
Run the code above in your browser using DataLab