predQuantile: Analyze predictions of supervised model in quantiles

In brooksandrew/Rsenal: Rsenal (arsenal) of assorted R functions developed over the years

Description

Analyzing results of supervised model using test data and quantiles. Note only model or testPred are necessary, not both. testPred is advised.

Usage

predQuantile(model = NULL, xtest = NULL, ytest, n = 5, roundText = 3,
  testPred = NULL, fw = NULL)

Arguments

model	the model to get predictions from. Uses predict method. If testPred is specified, we don't need this parameter, Probably safer to just use testPred, unless the model you have has a predict method that will work as intended without any other arguments.
ytest	vector of test data (target variable)
n	number of quantiles
roundText	decimals to print
testPred	vector of predictions from test data. Do not use if using model parameter.
fw	numeric vector, of bins. Defaults to
xtext	data.frame of test data (predictor variables only)

Value

data.frame of results

See Also

predSortPlot

Examples

require('randomForest')
rf <- randomForest(x=mtcars[1:25,1:8], y=as.factor(mtcars[1:25, 'am']), ntree=5)
mtcarsTestPred <- predict(rf, mtcars[26:32, 1:8], type='prob')[,2]
pq2 <- predQuantile(xtest=mtcars[26:32, 1:8], ytest=mtcars[26:32, 'am'], n=3, roundText=4, testPred=mtcarsTestPred)
pq <- predQuantile(xtest=mtcars[26:32, 1:8], ytest=mtcars[26:32, 'am'], fw=seq(.1,1,length.out=10), roundText=4, testPred=mtcarsTestPred)

barplot(pq$actual, names=pq$predRange, ylab='True Positives', las=2, cex.names=.7)
barplot(pq$hitRate, names=pq$predRange, ylab='True Positive Hit Rate', las=2, cex.names=.7)
barplot(pq$predMax-pq$predMin, pq$hitRate, names=pq$predRange, ylab='prediction', las=2, cex.names=.7)

brooksandrew/Rsenal documentation built on May 13, 2019, 7:50 a.m.