BTYD (Buy 'Till You Die). Walkthrough. pnbd.EstimateParameters()

问题

This is the first time I am working on BTYD procedure. I am having errors running the parameter estimates. I have provided the error message below. I have been following the BTYD - Walkthrough.

Does anybody know how to fix this? I worked through the sample data set and it works fine. I uploaded my file in the same format, it wouldn't work. There are no missing or empty rows/values. Help would be greatly appreciated!!!

end.of.cal.period <- as.Date("2013-08-18")
elog.cal <- elog[which(elog$date <= end.of.cal.period), ]

split.data <- dc.SplitUpElogForRepeatTrans(elog.cal);
clean.elog <- split.data$repeat.trans.elog; 
#clean.elog
freq.cbt <- dc.CreateFreqCBT(clean.elog); 
freq.cbt[1:3,1:5]

tot.cbt <- dc.CreateFreqCBT(elog.cal) #used elog.cal instead of elog
cal.cbt <- dc.MergeCustomers(tot.cbt, freq.cbt)

birth.periods <- split.data$cust.data$birth.per
last.dates <- split.data$cust.data$last.date
cal.cbs.dates <- data.frame(birth.periods, last.dates, end.of.cal.period)
cal.cbs <- dc.BuildCBSFromCBTAndDates(cal.cbt, cal.cbs.dates,per="week") #works. no errors
head(cal.cbs, n=10)

sel <- cbind(cal.cbs)
colnames(sel) <-c('x', 't.x', 'T.cal')
head(sel, n=10)

params <- pnbd.EstimateParameters(sel)

Error in optim(logparams, pnbd.eLL, cal.cbs = cal.cbs, max.param.value = max.param.value, : L-BFGS-B needs finite values of 'fn'

回答1:

It turns out that when estimating the parameters, large values in sel$x will cause the exponentiation to blow up and error.

This guy has a fix on github here: https://github.com/theofilos/BTYD

I basically took all of his code in pnbd.R, and prepend it to my code for the Pareto/NBD analysis and it seems to fix that issue.

回答2:

Have you tried adjusting the starting parameters? They default to 1,1,1,1. Something like:

startingparams <- c(.1, 3, .1, .1)
params <- pnbd.EstimateParameters(sel, startingparams)

Play with different orders of magnitude and see if any stick. Alternatively, check your maximum 'x' value in the cbs matrix. If there are extreme outliers you may want to remove them and try estimating the parameters again.

来源：https://stackoverflow.com/questions/26429008/btyd-buy-till-you-die-walkthrough-pnbd-estimateparameters

标签

statistics