R Package for Pairwise Preference Elicitation

Allows users to derive multi-objective weights from pairwise comparisons, which research shows is more repeatable, transparent, and intuitive other techniques. These weights can be rank existing alternatives or to define a multi-objective utility function for optimization.


Build Status codecov.io

What is preference elicitation?

Most real-world decisions must reconcile multiple, competing objectives. In buying a car, you might be concerned about cost, reliability, and performance, but before you can make a decision, you must establish the relative importance of these goals. A common mathematical approach to this problem is to define weights for each of these objectives. Although you might have a ballpark intuition for the weights, it is difficult to set them in a repeatable and defendable manner.

Preference elicitation relieves some of this burden. Instead of determining the weights directly, you make a series of pairwise comparisons between alternatives: do you prefer car A, car B, or are you indifferent? Research has shown that these pairwise comparisons are far easier to make and much easier to justify than explicitly setting the weights directly. This package implements a preference elicitation algorithm that takes your stated preferences and uses them to calculate an optimal set of weights. It can even suggest which comparisons you should make to get the most accurate weights with the fewest number of queries.

Technical details about how this package works can be found in the article here.

Installation

This package is not yet added to CRAN, but you can install it directly from Github with the devtools package.

library(devtools)
install_github("jlepird/prefeR")

Examples

Hello, World

library(prefeR)
 
# Each column of data is a variable, i.e. objective, 
# and each row is an alternative.
p <- prefEl(data = data.frame(x = c(1, 0, 1), 
                              y = c(0, 1, 1)))
# Set the prior belief on the weights for objectives x and y
p$priors <- c(Normal(0, 1), 
              Normal(0, 1))
 
# Add in some pairwise preferences
p$addPref(1 %>% 3) # prefer row 1 to row 3
p$addPref(3 %<% 2) # prefer row 2 to row 3
 
# Run the inference
p$infer()   # returns that x and y are of equal importance
 
# What comparision should you make next?
p$suggest() # suggest compare 1 to 2
p$addPref(1 %=% 2)
 
# Re-run the infernence algorithm 
p$infer()   # maintains belief that 1 and 2 are equal
p$rank()    # calculates the value of all three alternatives

Choosing the best car in the Motor Trends dataset

library(prefeR)
vignette("mtcars")

News

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("prefeR")

0.1.1 by John Lepird, 2 years ago


Browse source code at https://github.com/cran/prefeR


Authors: John Lepird


Documentation:   PDF Manual  


MIT + file LICENSE license


Imports mcmc, methods, entropy

Suggests testthat, knitr, rmarkdown


See at CRAN