getChanges: Get change in prediction for an updated feature.

In rfFC: Random Forest Feature Contributions

Description

This method calculates the changes in predictions, for a pre-determined Random Forest model and set of instances, resulting from updating the value(s) of a specified (vector of) feature(s). The method works with regression and classification models. In case of binary classification the predictions are calculated by predictBC() or predict.randomForest and represent the probabilities of being in a given class. If the model was obtained directly via the randomForest() function, the type of predictions calculated correspond to predict.randomForest(). However, if the model is a binary classification model and was obtained via post-processing the original model from randomForest(), using prepareForPredictBC(), the type of predictions calculated correspond to predictBC().

Usage

getChanges(features, dataT, object, value=NULL, type=NULL, mcls=NULL)

Arguments

features	a vector of the feature numbers/names to be updated
dataT	a data frame containing the variables in the model for all instances for which changes in predictions are desired
object	an object of the class randomForest
value	a vector of new feature values for the features provided in features N.B. If this is set to NULL, and the specified features are binary, the prediction changes reported are those associated with the only possible change in value for these features: from 1 to 0 or vice-versa.
type	the type of the predictions considered for classification models, by default it is set to type="prob" but can be set to type="votes".
mcls	main class that be set to "1" for binary classification. If NULL, the class name from the first record in dataT will be set as "1"

Value

A matrix n x m of prediction changes, n is the number of instances in dataT and m is the number of updated features.

Author(s)

Anna Palczewska [e-mail] and
Richard Marchese Robinson [e-mail]

See Also

randomForest

Examples

library(randomForest)
data(ames)
ames_train<-ames[ames$Type=="Train",-c(1,3, ncol(ames))]
ames_train<-ames_train[1:100,]
rF_Model <- randomForest(x=ames_train[,-1],y=as.factor(as.character(ames_train[,1])),
  ntree=500,importance=TRUE, keep.inbag=TRUE,replace=FALSE) 
gc <- getChanges(c(1,166), ames_train, rF_Model)
change<-getChanges(c(1), ames_train[1, ], rF_Model, value = c(0.49))

rfFC documentation built on May 2, 2019, 5:18 p.m.