semipadd2pop-package: Combine two data sets to fit semiparametric additive models

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

Combine two data sets to fit semiparametric additive models

Description

Fit semiparametric additive models for linear and logistic regression as well as logistic regression with group testing data using sparsity and smoothness penalization. Allows for combining two data sets which share some common covariates in the fitted of the common effects.

Details

The principal functions in the package for fitting sparse semiparametric regression models on single data set are semipadd and semipadd_cv_adapt. The first requires the user to choose the sparsity tuning parameter value, while the second chooses it via crossvalidation and includes an adaptive step. The functions semipadd2pop and semipadd2pop_cv_adapt are for fitting sparse semiparametric regression models on two data sets which have some covariates in common, such that the fitted effects of the common covariates are encouraged via a ridge-like penalty to be similar. The first requires the user to choose all tuning parameters, while the second selects the tuning parameters governing sparsity and penalization towards similarity via crossvalidation, using an adaptive step.

Author(s)

Karl Gregory

Maintainer: Karl Gregory <[e-mail]>

References

This is a package for a working paper with Chris McMahan and Dewei Wang entitled “Combining data sets for semiparametric additive modeling”.

Examples

  ## Not run: 
# fit a semiparametric model on a single simulated data set
data <- get_semipadd_data(n = 500, response = "continuous")

semipadd.out <- semipadd(Y = data$Y,
                         X = data$X,
                         nonparm = data$nonparm,
                         response = "continuous",
                          w = 1,
                          d = 20,
                          xi = 1,
                          lambda.beta = 1,
                          lambda.f = 1,
                          tol = 1e-3,
                          max.iter = 500)
 
plot_semipadd(semipadd.out, 
              true.functions = list( f = data$f,
                                     X = data$X))

# fit semiparametric models combining two simulated data sets with some common covariates                                    
data <- get_semipadd2pop_data(n1 = 500, n2 = 600, response = "binary")

semipadd2pop.out <- semipadd2pop(Y1 = data$Y1,
                                 X1 = data$X1,
                                 nonparm1 = data$nonparm1,
                                 Y2 = data$Y2,
                                 X2 = data$X2,
                                 nonparm2 = data$nonparm2,
                                 response = "binary",
                                 rho1 = 1,
                                 rho2 = 1,
                                 nCom = data$nCom,
                                 d1 = 25,
                                 d2 = 15,
                                 xi = .5,
                                 w1 = 1,
                                 w2 = 1,
                                 w = 1,
                                 lambda.beta = .01,
                                 lambda.f = .01,
                                 eta.beta = .1,
                                 eta.f = .1,
                                 tol = 1e-3,
                                 maxiter = 500)
                             
plot_semipadd2pop(semipadd2pop.out,
                  true.functions = list(f1 = data$f1,
                                        f2 = data$f2,
                                        X1 = data$X1,
                                        X2 = data$X2))
               
  
## End(Not run)

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