LidarLDA: Fit LDA model to LIDAR data

In drvalle1/LidarLDA: Fits a Latent Dirichlet Allocation (LDA) model to Lidar data

Fit LDA model to LIDAR data

Description

This function uses a Gibbs sampler to fit a LDA model to LIDAR data. These data consist of two P x H matrices (matrices y and n), where P is for pixel and H is for height class.

Usage

LidarLDA(
  y,
  n,
  nclust,
  a.phi,
  b.phi,
  gamma,
  ngibbs,
  nburn,
  theta.post,
  phi.post,
  theta.init = NULL,
  phi.init = NULL
)

Arguments

y	P x H matrix containing the number of returns for each pixel in each height class
n	P x H matrix containing the number of incoming light pulses for each pixel in each height class
nclust	maximum number of clusters
a.phi	parameter > 0 for the prior of phi
b.phi	parameter > 0 for the prior of phi
gamma	parameter between 0 and 1 for the prior of the truncated stick-breaking prior
ngibbs	number of iterations for the MCMC algorithm
nburn	number of iterations to discard as burn in
theta.post	should samples from the posterior distribution for theta be returned (TRUE or FALSE)? If FALSE, just the posterior mean is returned
phi.post	should samples from the posterior distribution for phi be returned (TRUE or FALSE)? If FALSE, just the posterior mean is returned
theta.init	initial values, if available, for the P x nclust theta matrix. Default value for theta.init is NULL.
phi.init	initial values, if available, for the nclust x H phi matrix. Default value for phi.init is NULL.

Value

This function returns a list containing several elements:

• llk: log-likelihood for each iteration. This is a vector of size ngibbs.

• theta: estimated relative abundance of each cluster in each pixel. If theta.post==T, then this consists of a matrix where rows are samples from the posterior distribution. If theta.post==F, then this consists of a P x K matrix (K is the number of clusters and P is the number of pixels) containing the posterior mean.

• phi: estimated absorptance probability in each height class for each cluster. If phi.post==T, then this consists of a matrix where rows are samples from the posterior distribution. If phi.post==F, then this consists of a K x H matrix (K is the number of clusters and H is the number of height classes) containing the posterior mean.

drvalle1/LidarLDA documentation built on July 27, 2024, 7:24 a.m.