Skip to contents

This function runs BRS. It allows the user either to run BRS only once on the original data or to run BRS on bootstrapped samples

Usage

BRS(
  df,
  Y,
  maxLen,
  trainProp,
  numIter = 500L,
  numChain = 2L,
  numMine = 5000L,
  supp = 5L,
  alpha_1 = 50L,
  beta_1 = 1L,
  alpha_2 = 50L,
  beta_2 = 1L,
  prior_type = "beta",
  alpha_l = NULL,
  beta_l = NULL,
  lambda = NULL,
  nu = NULL,
  print = FALSE,
  bootstrap = FALSE,
  reps = 1L,
  sampleSize = NULL,
  seed = NULL
)

Arguments

df

dataframe of binary features

Y

vector of binary outcome

maxLen

integer maximum length of rules

trainProp

numeric for proportion of data to use as training data

numIter

integer for number of iterations in simulated annealing

numChain

integer for number of chains of simulated annealing

numMine

integer for number of rules to mine

supp

integer for percent minimum support

alpha_1

numeric for alpha_+ from the paper

beta_1

beta_+

alpha_2

alpha_-

beta_2

beta_-

prior_type

string for the prior type. Either "beta" for BRS-BetaBinomial or "poisson" for BRS-Poisson

alpha_l

vector of alpha_l for l=1...maxLen. If set to NULL and prior_type="beta", values will be automatically generated. Ignored if prior_type="poisson"

beta_l

vector of beta_l for l=1...maxLen. If set to NULL and prior_type="beta", values will be automatically generated. Ignored if prior_type="poisson"

lambda

numeric rate parameter for the prior on the number of rules

nu

numeric rate parameter for the prior on the length of rules

print

logical whether to print progress of algorithm

bootstrap

logical for whether to bootstrap

reps

integer for number of bootstrap reps

sampleSize

integer for bootstrap sample size. If set to NULL, default is the number of observations in data

Value

list of rule sets from all bootstrap replications

indices of bootstrap samples

accuracy, true positive rate, and false positive rate on test data for each rule set