trainCurveELBO: Training curve for Evidence Lower Bound (ELBO)

In MOFA: Multi-Omics Factor Analysis (MOFA)

Description

MOFA inference is done using the variational Bayes algorithm, which maximises a quantity called the Evidence Lower Bound (ELBO). The ELBO is supposed to increase monotonically up to convergence, but it can decrease substantially when dropping inactive factors. For more details read the supplementary methods. The frequency of ELBO computation as well as the convergence criteria are defined as hyperparameters in prepareMOFA.
All Training statistics, including the ELBO, can be fetch from the TrainStats slot of MOFAmodel .

Usage

trainCurveELBO(object, logScale = FALSE)

Arguments

object	a MOFAmodel object.
logScale	boolean indicating whether to apply log transform

Value

plot of ELBO values during training

Examples

# Example on the CLL data
filepath <- system.file("extdata", "CLL_model.hdf5", package = "MOFAdata")
MOFA_CLL <- loadModel(filepath)
trainCurveELBO(MOFA_CLL)
trainCurveELBO(MOFA_CLL, logScale= TRUE)

# Example on the scMT data
filepath <- system.file("extdata", "scMT_model.hdf5", package = "MOFAdata")
MOFA_scMT <- loadModel(filepath)
trainCurveELBO(MOFA_scMT)

MOFA documentation built on Feb. 11, 2021, 2:01 a.m.