pBARTr: Fit a bayesian additive regression tree(BART) model for...

In DongyueXie/bCART: Explore variants of Bayesian trees.

Description

Fit a bayesian additive regression tree(BART) model for classification

Usage

pBARTr(X, y, x.test, cutoff = 0.5, k = 2, binaryOffset = NULL,
  power = 2, base = 0.95, p_split = "CGM", r = 2, ntree = 50,
  ndpost = 700, nskip = 300, Tmin = 2, printevery = 100,
  p_modify = c(0.5, 0.5, 0), save_trees = F, rule = "bart",
  pre_train = F, n_pre_train = 100)

Arguments

X	training samples by features matrix
y	response, factor with level 0,1
x.test	testing samples by feature matrix
cutoff	label = 1 if p>cutoff; else label = 0.
k, power, base	see ?BART::pbart
binaryOffset	The model is P(Y=1 | x) = F(f(x) + binaryOffset).
p_split	choice of 'CGM', 'RS'. 'CGM' splits an internal node with probability base*(1+d)^(-power); 'RS': r^(-d), 2 <= r <= n. d is the depth of an internal node.
r	'RS' splits an internal node with probability r^(-d), 2 <= r <= n. d is the depth of an internal node.
ntree	number of trees
nskip, ndpost	number of burn-in and posterior draws
Tmin	minimum number of samples in a leaf node allowed
printevery	print progress for every 'printevery' iterations
p_modify	proportion of three moves: grow, prune, change.
save_trees	whether save all the trees from each iteration as a list
rule	The splitting rule of an internal node. Choices are: 1. "grp": Gaussian random projection, randomly draw a length p vector from standard normal as the linear combination coefficients of p variables; 2. sgrp: sparse Gaussian random projection, which generates sparse linear combination coefficients; 3. bart: originla bart splits, which are axis-aligned splits; 4. hyperplane: randomly connect two points from the node as the partiton of node space.
pre_train	whether pre-train the BART model using 'bart' rule before switching to another splitting rule.
n_pre_train	number of iterations of pre-train

Value

BARTr returns a list of the following elements.

yhat.train	A matrix with ndpost rows and nrow(X) columns.
yhat.test	A matrix with ndpost rows and nrow(x.test) columns.
yhat.train.mean	Posterior mean of MCMC draws of traning data fits
yhat.test.mean	Posterior mean of MCMC draws of testing data fits
sigma	draws of random error vairaince, length = nskip+ndpost
tree_history	If save_trees = TRUE, then a list of all trees
tree_proposal_total	A ntree by length(p_modify) matrix, the (i,j)th entry is the total number of jth proposal of ith tree
tree_proposal_accept	A ntree by length(p_modify) matrix, the (i,j)th entry is the total number of accepted jth proposal of ith tree
tree_leaf_count	Number of leaf nodes in each tree

Author(s)

Dongyue Xie: [e-mail]

References

Chipman, H., George, E., and McCulloch R. (2010) Bayesian Additive Regression Trees. The Annals of Applied Statistics, 4,1, 266-298 <doi:10.1214/09-AOAS285>.

DongyueXie/bCART documentation built on Feb. 4, 2020, 12:26 a.m.