SVGD_bayesian_nn: Main Function

In BangyaoZhao/svgd: SVGD for Bayesian Neural Network

Description

Main Function

Usage

SVGD_bayesian_nn(
  X_train,
  y_train,
  eigenMat = diag(x = apply(y_train, 2, var)),
  X_test = NULL,
  y_test = NULL,
  dev_split = 0.1,
  M = 20,
  num_nodes = c(20, 1),
  a0 = 1,
  b0 = 0.1,
  initial_values = FALSE
)

Arguments

X_train	The training dataset variables, a matrix with rows representing observations and columns representing covariates.
y_train	The training dataset outcomes, a vector with the length same as the number of rows of 'X_train'.
eigenMat	the variance matrix of the outcome
X_test	The testing data set variables, a matrix with the same number of columns as 'X_train'.
y_test	The testing dataset outcomes, a vector with the length same as the number of rows of 'X_test'.
M	The number of particles.
num_nodes	The number of nodes in each hidden layer (does not include the last layer, because the node in the last layer is always 1).
a0	a0, for the prior distribution of lambda and gamma.
b0	b0, for the prior distribution of lambda and gamma.
batch_size	The batch size.
max_iter	The maximum number of iterations.
master_stepsize	The master stepsize, which is needed to adjust convergence if using adagrad for optimization of the NN.
auto_corr	The auto correlation, which is needed to adjust convergence if using adagrad for optimization of the NN.
method	The optimization method to be used.
use_autodiff	Whether to use autodiffr, default to FALSE.

Value

A list containing:

• theta: The estimated parameters in the vector format.

• scaling_coef: The scaling coefficient

• svgd_rmse: The RMSE on the training data

• svgd_11: log likelihood

BangyaoZhao/svgd documentation built on Sept. 20, 2021, 2:35 a.m.