softbart_regression: SoftBart Regression

In SoftBart: Implements the SoftBart Algorithm

SoftBart Regression

Description

Fits a semiparametric regression model with the nonparametric function modeled using a SoftBart model.

Usage

softbart_regression(
  formula,
  data,
  test_data,
  num_tree = 20,
  k = 2,
  hypers = NULL,
  opts = NULL,
  verbose = TRUE
)

Arguments

formula	A model formula with a numeric variable on the left-hand-side and predictors on the right-hand-side.
data	A data frame consisting of the training data.
test_data	A data frame consisting of the testing data.
num_tree	The number of trees in the ensemble to use.
k	Determines the standard deviation of the leaf node parameters, which is given by 3 / k / sqrt(num_tree).
hypers	A list of hyperparameters constructed from the Hypers() function (num_tree, k, and sigma_mu are overridden by this function).
opts	A list of options for running the chain constructed from the Opts() function (update_sigma is overridden by this function).
verbose	If TRUE, progress of the chain will be printed to the console.

Value

Returns a list with the following components:

• sigma_mu: samples of the standard deviation of the leaf node parameters.

• sigma: samples of the error standard deviation.

• var_counts: a matrix with a column for each predictor group containing the number of times each predictor is used in the ensemble at each iteration.

• mu_train: samples of the nonparametric function evaluated on the training set.

• mu_test: samples of the nonparametric function evaluated on the test set.

• mu_train_mean: posterior mean of mu_train.

• mu_test_mean: posterior mean of mu_test.

• formula: the formula specified by the user.

• ecdfs: empirical distribution functions, used by the predict function.

• opts: the options used when running the chain.

• mu_Y, sd_Y: used with the predict function to transform predictions.

• forest: a forest object; see the MakeForest documentation for more details.

Examples

## NOTE: SET NUMBER OF BURN IN AND SAMPLE ITERATIONS HIGHER IN PRACTICE

num_burn <- 10 ## Should be ~ 5000
num_save <- 10 ## Should be ~ 5000

set.seed(1234)
f_fried <- function(x) 10 * sin(pi * x[,1] * x[,2]) + 20 * (x[,3] - 0.5)^2 + 
  10 * x[,4] + 5 * x[,5]

gen_data <- function(n_train, n_test, P, sigma) {
  X <- matrix(runif(n_train * P), nrow = n_train)
  mu <- f_fried(X)
  X_test <- matrix(runif(n_test * P), nrow = n_test)
  mu_test <- f_fried(X_test)
  Y <- mu + sigma * rnorm(n_train)
  Y_test <- mu + sigma * rnorm(n_test)
  
  return(list(X = X, Y = Y, mu = mu, X_test = X_test, Y_test = Y_test, 
              mu_test = mu_test))
}

## Simiulate dataset
sim_data <- gen_data(250, 250, 100, 1)

df <- data.frame(X = sim_data$X, Y = sim_data$Y)
df_test <- data.frame(X = sim_data$X_test, Y = sim_data$Y_test)

## Fit the model

opts <- Opts(num_burn = num_burn, num_save = num_save)
fitted_reg <- softbart_regression(Y ~ ., df, df_test, opts = opts)

## Plot results

plot(colMeans(fitted_reg$mu_test), sim_data$mu_test)
abline(a = 0, b = 1)

SoftBart documentation built on June 8, 2025, 9:40 p.m.