semipadd2pop_cv_adapt: Compute semiparametric binary-response regression model with...

In gregorkb/semipadd2pop: Combine two data sets to fit semiparametric additive models

Compute semiparametric binary-response regression model with 2 data sets while penalizing dissimilarity using CV to select tuning parameters after an adaptive step

Description

Compute semiparametric binary-response regression model with 2 data sets while penalizing dissimilarity using CV to select tuning parameters after an adaptive step

Usage

semipadd2pop_cv_adapt(
  Y1,
  X1,
  nonparm1,
  Y2,
  X2,
  nonparm2,
  response,
  rho1,
  rho2,
  w1 = 1,
  w2 = 1,
  w = 1,
  nCom,
  d1,
  d2,
  xi,
  n.lambda = 5,
  n.eta = 5,
  lambda.min.ratio = 0.01,
  lambda.max.ratio = 1,
  eta.min.ratio = 0.001,
  eta.max.ratio = 10,
  n.folds = 5,
  lambda.beta = 1,
  lambda.f = 1,
  eta.beta = 1,
  eta.f = 1,
  tol = 0.001,
  maxiter = 1000,
  report.prog = FALSE
)

Arguments

Y1	the response data for data set 1
X1	the matrix with the observed covariate values for data set 1 (including a column of ones for the intercept)
nonparm1	a vector indicating for which covariates a nonparametric function is to be estimated for data set 1
Y2	the response data for data set 2
X2	the matrix with the observed covariate values for data set 2 (including a column of ones for the intercept)
nonparm2	a vector indicating for which covariates a nonparametric function is to be estimated for data set 2
response	a character string indicating the type of response. Can be "continuous", "binary", or "gt"
rho1	weight placed on the first data set
rho2	weight placed on the second data set
w1	covariate-specific weights for different penalization among covariates in data set 1
w2	covariate-specific weights for different penalization among covariates in data set 2
w	covariate-specific weights for different penalization toward similarity for different covariates
nCom	the number of covariates to be treated as common between the two data sets: these must be arranged in the first nCom columns of the matrices X1 and X2 after the column of ones corresponding to the intercept.
d1	vector giving the dimensions the B-spline bases to be used when fitting the nonparametric effects. If a scalar is given, this dimension is used for all nonparametric effects.
d2	vector giving the dimensions the B-spline bases to be used when fitting the nonparametric effects. If a scalar is given, this dimension is used for all nonparametric effects.
xi	a tuning parameter governing the smoothness of the nonparametric estimates
n.lambda	the number of lambda values with which to make the grid
n.eta	the number of eta values with which to make the grid
lambda.min.ratio	ratio of the smallest lambda value to the smallest value of lambda which admits no variables to the model
lambda.max.ratio	ratio of the largest lambda value to the smallest value of lambda which admits no variables to the model
eta.min.ratio	ratio of the smallest to largest value in the sequence of eta values
eta.max.ratio	controls the largest value of eta in the eta sequence
n.folds	the number of crossvalidation folds
lambda.beta	the level of sparsity penalization for the parametric effects (relative to nonparametric effects)
lambda.f	the level of sparsity penalization for the nonparametric effects (relative to the parametric effects)
eta.beta	the level of penalization towards model similarity for parametric effects indicated to be common (relative to nonparametric effects)
eta.f	the level of penalization towards model similarity for nonparametric effects indicated to be common (relative to the parametric effects)
tol	a convergence criterion
maxiter	the maximum allowed number of iterations
report.prog	a logical indicating whether the progress of the algorithm should be printed to the console

Value

Returns the estimator of the semiparametric additive model

Examples

data <- get_semipadd2pop_data(n1 = 500,n2 = 300,response = "continuous")

semipadd2pop_cv_adapt.out <- semipadd2pop_cv_adapt(Y1 = data$Y1,
                                                   X1 = data$X1,
                                                   nonparm1 = data$nonparm1,
                                                   Y2 = data$Y2,
                                                   X2 = data$X2,
                                                   nonparm2 = data$nonparm2,
                                                   response = "continuous",
                                                   rho1 = 2,
                                                   rho2 = 1,
                                                   w1 = 1,
                                                   w2 = 1,
                                                   w = 1,
                                                   nCom = data$nCom,
                                                   d1 = 25,
                                                   d2 = 15,
                                                   xi = .5,
                                                   n.lambda = 5,
                                                   n.eta = 5,
                                                   lambda.min.ratio = 0.01,
                                                   lambda.max.ratio = 0.10,
                                                   n.folds = 5,
                                                   lambda.beta = 1,
                                                   lambda.f = 1,
                                                   eta.beta = 1,
                                                   eta.f = 1,
                                                   tol = 1e-3,
                                                   maxiter = 1000,
                                                   report.prog = TRUE)

plot_semipadd2pop_cv_adapt(semipadd2pop_cv_adapt.out,
                           true.functions = list(f1 = data$f1,
                                                 f2 = data$f2,
                                                 X1 = data$X1,
                                                 X2 = data$X2))

gregorkb/semipadd2pop documentation built on Oct. 2, 2022, 1:37 p.m.