estimateRegularity: Estimate Regularity in Intertransaction Timings

Description

Estimates degree of regularity of intertransaction timings of a customer cohort,

Usage

estimateRegularity(elog, method = "wheat", plot = FALSE)

Arguments

elog

data.frame with transaction logs; requires columns customer-id 'cust' and transaction time t or 'date'

method

'wheat', 'mle', 'mle-minka', 'mle-thom', 'cv'

plot

if TRUE then distribution of estimated regularity will be plotted

Value

estimated real-valued regularity parameter; rounded to an integer, this can be used as 'k' for estimating CBG/CNBD-k models

Details

This is done - by either assuming same regularity across all customers; this then only require three transactions per customer method: wheat - or by estimating regularity for each customer seperately (as the shape parameter of a fitted gamma distribution), and then return the median across estimates; this requires min. 10 transactions per customer methods: mle, mle-minka, mle-thom, cv

Examples

Run this code

# NOT RUN {
# generate Erlang-3 with various lambdas
nr_of_customers <- 100
elog <- rbindlist(lapply(1:nr_of_customers, function(i) {
  k <- 3
  lambda <- exp(rnorm(1))
  data.table(cust=i, t=cumsum(rgamma(20, k, k*lambda)))
}))

# estimate regularity parameter k
estimateRegularity(elog, plot=T, method='wheat')
estimateRegularity(elog, plot=T, method='mle-minka')
estimateRegularity(elog, plot=T, method='mle-thom')
estimateRegularity(elog, plot=T, method='cv')
# }