Create a multi-armed Bayesian bandit object.

Source: R/bandit.R

Fit a multi-armed bandit object based on a bayesTest which can serve recommendations and adapt to new data.

banditize(bT, param, higher_is_better = TRUE)

Arguments

bT

a bayesTest object
param

which model parameter (posterior) to evaluate; defaults to first param
higher_is_better

is a higher value of `param` equivalent to a better choice?

Value

A bayesBandit object.

Details

banditize is an 'object-oriented' implementation of multi-armed bandits in bayesAB. It is useful in conjunction with a Shiny app or Plumber deployment. The object itself is mutable and can adapt/learn from new data without having to re-assign the variable.

Comes with 5 methods:

  • serveRecipe(): serves a recipe to show your user based on samples from both posteriors.

  • setResults(results): set results for one or more recipes for one or more instances of feedback. Used to update bandit.

  • getBayesTest(): returns most updated bayesTest object.

  • getOriginalTest(): returns original bayesTest object without any updates.

  • getUpdates(): returns a summarized version of all updates this bandit has processed.

Examples

A_binom <- rbinom(100, 1, .5)
B_binom <- rbinom(100, 1, .6)

AB1 <- bayesTest(A_binom, B_binom, priors = c('alpha' = 1, 'beta' = 1), distribution = 'bernoulli')

binomialBandit <- banditize(AB1)
binomialBandit$serveRecipe()

#> [1] "B"
binomialBandit$setResults(list('A' = c(1, 0, 1, 0, 0), 'B' = c(0, 0, 0, 0, 1)))

#> [1] 0