tensorBF (version 1.0.2)

getDefaultOpts: A function for generating a default set of parameters for Bayesian Tensor Factorization methods

Description

getDefaultOpts returns the default choices for model parameters.

Usage

getDefaultOpts(method = "CP")

Arguments

method

the factorization method for which options are required. Currently only "CP" (default) is supported.

Value

A list with the following model options:

ARDX

TRUE: use elementwise ARD prior for X, resulting in sparse X's. FALSE: use guassian prior for a dense X (default).

ARDW

TRUE: use elementwise ARD prior for W, resulting in sparse W's (default). FALSE: use guassian prior for a dense W.

ARDU

TRUE: use elementwise ARD prior for U, resulting in sparse U's. FALSE: use guassian prior for a dense U (default).

iter.burnin

The number of burn-in samples (default 5000).

iter.sampling

The number of saved posterior samples (default 50).

iter.thinning

The thinning factor to use in saving posterior samples (default 10).

prior.alpha_0t

The shape parameter for residual noise (tau's) prior (default 1).

prior.beta_0t

The rate parameter for residual noise (tau's) prior (default 1).

prior.alpha_0

The shape parameter for the ARD precisions (default 1e-3).

prior.beta_0

The rate parameter for the ARD precisions (default 1e-3).

prior.betaW1

Bernoulli prior for component activiations, prior.betaW1 < prior.betaW2: sparsity inducing (default: 1).

prior.betaW2

Bernoulli prior for component activation, (default: 1).

init.tau

The initial value for noise precision (default 1e3).

verbose

The verbosity level. 0=no printing, 1=moderate printing, 2=maximal printing (default 1).

checkConvergence

Check for the convergence of the data reconstruction, based on the Geweke diagnostic (default TRUE).

Details

This function returns options for defining the model's high-level structure (sparsity priors), the hyperparameters, and the uninformative priors. We recommend keeping these as provided.

Examples

# NOT RUN {
#To run the algorithm with other values:
opts <- getDefaultOpts()
opts$ARDW <- FALSE #Switch off Feature-level Sparsity on W's
 
# }
# NOT RUN {
res <- tensorBF(Y=Y,opts=opts)
# }