GFA (version 1.0.5)

informativeNoisePrior: Informative noise residual prior

Description

informativeNoisePrior returns an informative noise prior for GFA, for a given data collection. The function sets the noise residual parameters such that the expected proportion of variance explained is for all variables and groups (in contrast to being proportional to their original scale). Recommended e.g. when the data is 'small n, large p', and the standard prior from getDefaultOpts seems to overfit the model by not shutting off any component with high initial K.

Usage

informativeNoisePrior(Y, opts, noiseProportion = 0.5, conf = 1)

Value

The input model options (opts) with an updated residual noise prior, corresponding to the elements prior.alpha_0t and prior.beta_0t.

Arguments

Y

The data. For details, see function gfa.

opts

Model options. See function getDefaultOpts for details. If option tauGrouped is TRUE (default), each data source is given equal importance (feature importance may vary within each source). If it is FALSE, each feature is given equal importance.

noiseProportion

proportion of total variance to be explained by noise. Suggested to lie between 0.01 and 0.99.

conf

Confidence in the prior, relative to confidence in the data. Suggested to lie between 0.01 and 100.

Examples

#Given data collection Y
opts <- getDefaultOpts()
if (FALSE) opts <- informativeNoisePrior(Y,opts,0.2,1)
if (FALSE) res <- gfa(Y,opts=opts)