cont_selec: contracts explanatory variables, conducts model selection and...

In yorickuser/contselec: Model selection procesure by contracting explanatry variables, with check of each explanatry variable's strength.

Description

This function contracts explanatory variables and conduct best subset model selection, permutation test and stepwise model selection.

Usage

cont_selec(data00, pos_x = NULL, pos_y = NULL, target = 1,
  edge_cor = 0.52, edge_explain = 0.65, edge_param_number = 3,
  repeatn = 100, family = "glmm_poisson", check_spa_cor = F,
  use_pforeach = TRUE, perm = TRUE, nrep = 10000)

Arguments

data00	data.frame : an explained variable and explanatory variables. The explained variable is specified by "target".
pos_x	array of numeric : x-coordinates of spatial position for each sampled data in "data00".
pos_y	array of numeric : y-coordinates of spatial position for each sampled data in "data00".
target	integer or character : spefifies explained variables by its column id (integer) or column name (character).
edge_cor	real number : threshold for grouping; variables of correlation no less than this threshold are grouped together.
edge_explain	real number : threshold for the variance ratio to be explained, based on which the number of principal components "np" to represent the contracted group is determined.
edge_param_number	real number: this parameter constrains the number of explanatory variables in the subset model selection and stepwise model selection, so that [sample size]/[number of free parameter] >= "edge_param_number".
repeatn	integer : the number of model evaluations proccessed as a single job for each CPU core (This parameter is meaningful only when "use_pforeach" is TRUE).
family	character : error distribution. Possible choices are "glmm_poisson", "poisson","gaussian".
check_spa_cor	TRUE/FALSE : if TRUE, the best model with residuals significantly correlated with c(pos_x, pos_y) are removed repeatedly until the correlation becomes non-significant.
use_pforeach	TRUE/FALSE : if TRUE, pforeach is used instead of foreach in best subset selection and permutation tests (when "perm" is TRUE).
perm	TRUE/FALSE : if TRUE, permutation tests for explained variables are conducted.
nrep	integer : number of resampling in the permutation test.

Details

Spatial correlation of residuals of each fitted model can also be examimed.

Value

list(target=target, result=result, data0=data0, data1=data1, group=group, family=family, bestmodels=res_sb$bestmodels, bestmodel_stepwise=res_ss$bestmodel, bestmodel=res_sb$bestmodel, pos_x=pos_x, pos_y=pos_y)

Author(s)

Hiroshi C. Ito

Examples

data(Cars93, package = "MASS");
data=Cars93;
data=data[complete.cases(data),];
data=data[,sapply(data[1,],is.numeric)];

res=cont_selec(data,target="Horsepower",edge_cor=0.9,edge_explain=0.6,
               edge_param_number=3,family="gaussian",use_pforeach=FALSE);

plot_each_effect(res)
plot_combined_effect(res,sign_effect=-1)
plot_combined_effect(res,sign_effect=1)

print(res);
summary(res);

yorickuser/contselec documentation built on July 25, 2021, 8:14 a.m.