Buy Till You Die(BTYD) - Model Validation in R I have used the BTYD Package in R to estimate the number of transactions that a customer is expected to make in the future. I have 2 questions:
1) Is the number of transactions always a non-integer value?
2) How do you validate the performance of the BTYD Model. I have made some R plots using the functions pnbd.PlotFrequencyInCalibration, pnbd.PlotFreqVsConditionalExpectedFrequency and pnbd.PlotRecVsConditionalExpectedFrequency. What exactly is the method used to validate the performance of BTYD Model? Can these plots be considered a measure of checking the performance of the model? 
Any help will be appreciated.
 A: 1) The number of predicted transactions will either be an estimated # of transactions or a probabilistic value as a function of the form of the model. See the comments below for some clarification of how different models work and develop predictions.
2) Validation is dealt with in this paper on the BTYD model... https://cran.r-project.org/web/packages/BTYD/vignettes/BTYD-walkthrough.pdf Here's a quote (p. 10): 

To validate that the model works, we need to divide the data up into a
  calibration period and a holdout period.

The subsequent discussion walks through this issue quite thoroughly.
A: I’ve been using the library. The y axis is the mean expected transaction count for the model (mean of decimals), and the mean transaction count for the actual (mean of integers, since these are known counts). I agree, it makes no sense to only predict integers. You can just round them off if you like, but beware of introducing a cross sectional bias. If they’re mostly small positive values (like 0.1) rounding them all down will introduce a negative bias. Another solution: round up or down randomly, according to the fractional part.
