WGCNA (version 1.25-1)

votingLinearPredictor: Voting linear predictor

Description

Predictor based on univariate regression on all or selected given features that pools all predictions using weights derived from the univariate linear models.

Usage

votingLinearPredictor(
         x, y, xtest = NULL, 
         classify = FALSE, 
         CVfold = 0, 
         randomSeed = 12345, 
         assocFnc = "cor", assocOptions = "use = 'p'", 
         featureWeightPowers = NULL, priorWeights = NULL, 
         weighByPrediction = 0, 
         nFeatures.hi = NULL, nFeatures.lo = NULL, 
         dropUnusedDimensions = TRUE, 
         verbose = 2, indent = 0)

Arguments

x

Training features (predictive variables). Each column corresponds to a feature and each row to an observation.

y

The response variable. Can be a single vector or a matrix with arbitrary many columns. Number of rows (observations) must equal to the number of rows (observations) in x.

xtest

Optional test set data. A matrix of the same number of columns (i.e., features) as x. If test set data are not given, only the prediction on training data will be returned.

classify

Should the response be treated as a categorical variable? Classification really only works with two classes. (The function will run for multiclass problems as well, but the results will be sub-optimal.)

CVfold

Optional specification of cross-validation fold. If 0 (the default), no cross-validation is performed.

randomSeed

Random seed, used for observation selection for cross-validation. If NULL, the random generator is not reset.

assocFnc

Function to measure association. Usually a measure of correlation, for example Pearson correlation or bicor.

assocOptions

Character string specifying the options to be passed to the association function.

featureWeightPowers

Powers to which to raise the result of assocFnc to obtain weights. Can be a single number or a vector of arbitrary length; the returned value will contain one prediction per power.

priorWeights

Prior weights for the features. If given, must be either (1) a vector of the same length as the number of features (columns in x); (2) a matrix of dimensions length(featureWeightPowers)x(number of features); or (3) array of dimensions (number

weighByPrediction

(Optional) power to downweigh features that are not well predicted between training and test sets. See details.

nFeatures.hi

Optional restriction of the number of features to use. If given, this many features with the highest association and lowest association (if nFeatures.lo is not given) will be used for prediction.

nFeatures.lo

Optional restriction of the number of lowest (i.e., most negatively) associated features to use. Only used if nFeatures.hi is also non-NULL.

dropUnusedDimensions

Logical: should unused dimensions be dropped from the result?

verbose

Integer controling how verbose the diagnostic messages should be. Zero means silent.

indent

Indentation for the diagnostic messages. Zero means no indentation, each unit adds two spaces.

Value

• A list with the following components:
• predicted
  The back-substitution prediction on the training data. Normally an array of dimensions (number of observations) x (number of response variables) x length(featureWeightPowers), but unused are dropped unless dropUnusedDimensions = FALSE.
• weightBase
  Absolute value of the associations of each feature with each response.
• variableImportance
  The weight of each feature in the prediction (including the sign).
• predictedTest
  If input xtest is non-NULL, the predicted test response, in format analogous to predicted above.
• CVpredicted
  If input CVfold is non-zero, cross-validation prediction on the training data.

Details

The predictor calculates the association of each (selected) feature with the response and uses the association to calculate the weight of the feature as sign(association) * (association)^featureWeightPower. Optionally, this weight is multiplied by priorWeights. Further, a feature prediction weight can be used to downweigh features that are not well predicted by other features (see below).

For classification, the (continuous) result of the above calculation is turned into ordinal values essentially by rounding.

If features exhibit non-trivial correlations among themselves (such as, for example, in gene expression data), one can attempt to down-weigh features that do not exhibit the same correlation in the test set. This is done by using essentially the same predictor to predict _features_ from all other features in the test data (using the training data to train the feature predictor). Because test features are known, the prediction accuracy can be evaluated. If a feature is predicted badly (meaning the error in the test set is much larger than the error in the cross-validation prediction in training data), it may mean that its quality in the training or test data is low (for example, due to excessive noise or outliers). Such features can be downweighed using the argument weighByPrediction. The extra factor is min(1, (root mean square prediction error in test set)/(root mean square cross-validation prediction error in the trainig data)^weighByPrediction), that is it is never bigger than 1.

See Also

bicor for robust correlation that can be used as an association measure