OptimizeMetaFeatureCombo: Optimize the combination of predictors by metafeatures alone...

In ncats/MultiOmicsGraphPrediction: Network-Based Ensemble Prediction from IntLIM Predictors

Optimize the combination of predictors by metafeatures alone (in other words, exclude pooling and combine in a single layer using a linear combination).

Description

Optimize the combination of predictors by metafeatures alone (in other words, exclude pooling and combine in a single layer using a linear combination).

Usage

OptimizeMetaFeatureCombo(
  modelResults,
  verbose = TRUE,
  modelRetention = "stringent",
  useCutoff = FALSE,
  modelCountCutoff = 0,
  pruningTechnique = "backward.stepwise",
  stochastic = TRUE,
  doPooling = TRUE,
  doPruning = TRUE,
  averaging = FALSE,
  zeroOut = FALSE,
  feedback = FALSE,
  trimming = "modelwise"
)

Arguments

modelResults	An object of the ModelResults class.
verbose	Whether to print results as you run the model.
modelRetention	Strategy for model retention. "stringent" (the default) retains only models that improve the prediction score. "lenient" also retains models that neither improve nor reduce the prediction score.
useCutoff	Whether or not to use the cutoff for prediction. Default is FALSE.
modelCountCutoff	Only consider this number of models. If not provided, use all. Models will be selected by positive weight.
pruningTechnique	Pruning technique to use. Possible methods are "backward.stepwise", "forward.stepwise", "individual.performance", and "exhaustive".
stochastic	Whether to use a stochastic model (TRUE) or a batch model (FALSE)
doPooling	Whether or not to pool predictors together using the structure of the graph.
doPruning	Whether or not to prune predictors. If pruning is not done, result is a weighted combination of all predictors.
averaging	Whether to use averaging to combine predictors instead of retaining the same functional form for input and output.
zeroOut	This parameter zeros out predictors outside of the allowed range.
feedback	This parameter controls whether or not a feedback layer will be implemented, i.e. whether a full pruning procedure over the entire graph will be allowed to inform pruning of individual neighborhoods.
trimming	Set to "edgewise" to trim edges at each layer or "modelwise" to trim entire models (neighborhoods, connected components) at each layer.

Value

A ModelResults object, with all of the tracking information from each iteration filled in.

ncats/MultiOmicsGraphPrediction documentation built on Aug. 23, 2023, 9:19 a.m.