Usage
structureFinder(X = X, Z = NULL, Bic_null_vect = NULL, candidates = -1,
reject = 0, methode = 1, p1max = 5, p2max = NULL, Maxiter = 1,
plot = FALSE, best = TRUE, better = FALSE, random = TRUE,
verbose = 1, nb_opt_max = NULL, exact = TRUE, nbini = NULL,
star = TRUE, clean = TRUE, ...)
Arguments
Z
binary adjacency matrix of size p. if NULL zero matrix is used
Bic_null_vect
the BIC of the null hypothesis (used for independent variables)
candidates
0:row and column,-1:column only, int>0:random int candidates, -2 : all (but the diag), -3 : non-zeros
reject
0: constraint relaxation, 1: reject mode
methode
parameter for OLS (matrix inversion) 1:householderQr, 2:colPivHouseholderQr
p1max
maximum complexity for a regression
p2max
maximum number of regressions
plot
TRUE: returns for each step the type of move, complexity and BIC
best
TRUE: systematically jumps to the best BIC seen ever when seen (it is stored even if best=FALSE)
better
TRUE: systematically jumps to the best candidate if better than stationnarity (random wheighted jump otherwise)
random
if FALSE:moves only to improve and only to the best
verbose
0:none, 1:BIC,step and complexity when best BIC found 2:BIC, step, complexity, nb candidates and best candidate when best BIC found
nb_opt_max
stop criterion defining how many times the chain can walk (or stay) on the max found
exact
boolean. If exact subregression is found it gives its content (another verbose mode).
nbini
Number of initialisations (using Winitial if Z is NULL). if NULL and Zini is NULL : only one chain beginning with zero matrix.
star
boolean to compute BIC* instead of BIC (stronger penalization of the complexity). WARNING : star=TRUE implies p2max
clean
cleaning steps at the end of the walk (testing each remainging 1 for removal)
...
parameters to be passed (for Winitial).