modelGradient: Model log-likelihood/objective error function and its...

In ahonkela/tigre: Transcription factor Inference through Gaussian process Reconstruction of Expression

Description

modeGradient gives the gradient of the objective function for a model. By default the objective function (modelObjective) is a negative log likelihood (modelLogLikelihood).

Usage

  modelObjective(params, model, ...)
  modelLogLikelihood(model)
  modelGradient(params, model, ...)

Arguments

params	parameter vector to evaluate at.
model	model structure.
...	optional additional arguments.

Value

g	the gradient of the error function to be minimised.
v	the objective function value (lower is better).
ll	the log-likelihood value.

See Also

modelOptimise.

Examples

  # Load a mmgmos preprocessed fragment of the Drosophila developmental
  # time series
  data(drosophila_gpsim_fragment)

  # The probe identifier for TF 'twi'
  twi <- "143396_at"
  # The probe identifier for the target gene
  targetProbe <- "152715_at"

  # Create the model but do not optimise
  model <- GPLearn(drosophila_gpsim_fragment,
                   TF=twi, targets=targetProbe,
                   useGpdisim=TRUE, quiet=TRUE,
                   dontOptimise=TRUE)
  params <- modelExtractParam(model, only.values=FALSE)
  ll <- modelLogLikelihood(model)
  paramValues <- modelExtractParam(model, only.values=TRUE)
  modelGradient(paramValues, model)

ahonkela/tigre documentation built on Aug. 6, 2021, 12:08 p.m.