BC is the main function for running BayesComm models.
It is a wrapper function for BCfit; it checks inputs, sets up the model types and specifies a number of default BCfit settings.
BC(Y, X = NULL, model = "null", covlist = NULL, condition = NULL, its = 100, ...)BCfit
bayescomm containing the model call and parameter chains which can be viewed and manipulated using window, plot, print and summary.
Y must be a matrix with records as rows and species as columns and X a matrix with records as rows and covariates as columns.
model must be one of: "null" (intercept only), "environment" (intercept & covariates), "community" (intercept & community matrix) or "full" (intercept, covariates & community matrix).
covlist must have the same length as the number of species with, each element a vector of column indices for X. covlist defaults to NULL, which includes all covariates for all species.
For more details of arguments for model fitting see BCfit. condition is an optional matrix of conditioning variables.
These are fitted in the same way as X but are not removed in null and community models.
BCfit
# create fake data
n <- 100
nsp <- 4
k <- 3
X <- matrix(c(rep(1, n), rnorm(n * k)), n) # covariate matrix
W <- matrix(rnorm(nsp * nsp), nsp)
W <- W %*% t(W) / 2 # true covariance matrix
B <- matrix(rnorm(nsp * (k + 1), 0, 3), nsp) # true covariates
mu <- apply(B, 1, function(b, x) x %*% b, X) # true mean
e <- matrix(rnorm(n * nsp), n) %*% chol(W) # true e
z <- mu + e # true z
Y <- ifelse(z > 0, 1, 0) # true presence/absence
# run BC (after removing intercept column from design matrix)
m1 <- BC(Y, X[, -1], model = "full", its = 100)
Run the code above in your browser using DataLab