optimizeModelWeight: Optimize weight on ensemble of 2 supervised models

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

Description

This function creates a weighted average of predictions from two models and evaluates F1, precision, recall, auc or c for each combination of the models to determine the best weights for each.

Usage

optimizeModelWeight(pred1, pred2, actual, steps = 50, cutoff = 0.5)

Arguments

pred1	numeric vector of probabilities, prediction from model 1
pred2	numeric vector of probabilities, prediction from model 2
actual	vector of 1s and 0s. The target variable test data
steps	number, high numbers compute a more exhaustive combination of model weights
cutoff	Cutoff used to demarcate predictions into positive or negative class.
ytest,	vector of the target variable from test data. (0s and 1s)
jitterPlot	TRUE or FALSE. Jitters points on plot when TRUE

Value

data.frame of results

Examples

require('Hmisc')
fit_glm1 <- glm(am~cyl, data=mtcars, family=binomial(logit))
fit_glm2 <- glm(am~disp, data=mtcars, family=binomial(logit))
ow <- optimizeModelWeight(fit_glm1$fitted.values, fit_glm2$fitted.values, actual=fit_glm1$model$am)
plot(ow$weights, ow$precision, type='l', xlab='weight on model 1')

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