This program is used for imputing missing covariates by the 'sequential BART' approach. Sequential BART is a flexible Bayesian nonparametric approach to impute the missing covariates which involves factoring the joint distribution of the covariates with missingness into a set of sequential conditionals and applying Bayesian additive regression trees (BART) to model each of these univariate conditionals. Package provides a function, seqBART(), which computes and returns the imputed values.
devtools::install_github("mjdaniels/SequentialBART")
The pacakge provides a function, seqBART(), to run the sequential BART model to find the missing covariates. The function takes as arguments
-
x, is Covariates having the missing values.
-
y, is Response Variable.
-
x.type, is a vector indicating the type of covariates (0=binary, 1=continuous)
-
y.ype, is the type of response and the inference regression model used for imputation. It can take 5 values: y.type=0 for no response, y.type=1 for continuous response using linear regression for imputation, y.type=2 for binary response using logistic regression for imputation. Latest version
0.1.1has 2 new values fo y.type: 3 for continuous response using BART for imputation, 4 for binary response using BART probit for imputation. -
numimpute, is the Number of Imputed Datasets that will be generated. Default is = 5
-
seed_dist, is the value that will used to generate the distributions with. Default is = 12345
-
seed_draws, is the value that will used to generate the draws with. Default is = 99
Rest of the arguments are standard arguments for BART; Descriptions and defaults are provided in the pacakge help pages.
sbart::seqBART(x=Xcovariates, y=Response, x.type=datatypeValues, y.type=1)