mplatzer / btydplus Goto Github PK

View Code? Open in Web Editor NEW

182.0 35.0 49.0 951 KB

R package for Customer Behavior Analysis

Home Page: https://cran.r-project.org/web/packages/BTYDplus

License: GNU General Public License v3.0

R 87.15% C++ 11.09% TeX 1.77%

customer-behavior predictive-analytics crm marketing-science rstats

btydplus's Introduction

BTYDplus

The BTYDplus R package provides advanced statistical methods to describe and predict customer's purchase behavior. It uses historic transaction records to fit a probabilistic model, which then allows to compute quantities of managerial interest on a cohort- as well as on a customer level (Customer Lifetime Value, Customer Equity, P(alive), etc.).

This package complements the BTYD package by providing several additional buy-till-you-die models, that have been published in the marketing literature, but whose implementation are complex and non-trivial. These models are: NBD, MBG/NBD, BG/CNBD-k, MBG/CNBD-k, Pareto/NBD (HB), Pareto/NBD (Abe) and Pareto/GGG.

Installation

# install.packages("devtools")
devtools::install_github("mplatzer/BTYDplus", dependencies=TRUE)
library(BTYDplus)

Getting Started

demo("cdnow")        # Demonstration of fitting various models to the CDNow dataset
demo("mbg-cnbd-k")   # Demonstration of MBG/CNBD-k model with grocery dataset
demo("pareto-abe")   # Demonstration of Abe's Pareto/NBD variant with CDNow dataset
demo("pareto-ggg")   # Demonstration of Pareto/NBD (HB) & Pareto/GGG model with grocery dataset

Implemented Models

These R source files extend the functionality of the BTYD package by providing functions for parameter estimation and scoring for NBD, MBG/NBD, BG/CNBD-k, MBG/CNBD-k, Pareto/NBD (HB), Pareto/NBD (Abe) and Pareto/GGG.

NBD Ehrenberg, Andrew SC. "The pattern of consumer purchases." Applied Statistics (1959): 26-41. \doi{10.2307/2985810}
MBG/NBD Batislam, E.P., M. Denizel, A. Filiztekin. 2007. Empirical validation and comparison of models for customer base analysis. International Journal of Research in Marketing 24(3) 201–209. \doi{10.1016/j.ijresmar.2006.12.005}
(M)BG/CNBD-k Reutterer, T., Platzer, M., & Schroeder, N. (2020). "Leveraging purchase regularity for predicting customer behavior the easy way." International Journal of Research in Marketing. \doi{10.1016/j.ijresmar.2020.09.002}
Pareto/NBD (HB) Ma, Shao-Hui, and Jin-Lan Liu. "The MCMC approach for solving the Pareto/NBD model and possible extensions." Natural Computation, 2007. ICNC 2007. Third International Conference on. Vol. 2. IEEE, 2007. \doi{10.1109/ICNC.2007.728}
Pareto/NBD (Abe) Abe, Makoto. "Counting your customers one by one: A hierarchical Bayes extension to the Pareto/NBD model." Marketing Science 28.3 (2009): 541-553. \doi{10.1287/mksc.1090.0502}
Pareto/GGG Platzer, Michael, and Thomas Reutterer. "Ticking Away the Moments: Timing Regularity Helps to Better Predict Customer Activity." Marketing Science (2016). \doi{10.1287/mksc.2015.0963}

Contributions

We certainly welcome all feedback and contributions to this package! Please use GitHub Issues for filing bug reports and feature requests, and provide your contributions in the form of Pull Requests. See also these general guidelines to contribute to Open Source projects on GitHub.

btydplus's People

Contributors

Stargazers

Watchers

btydplus's Issues

Add Zhang's Clumpiness Measure

increase code coverage to 100%

test MBG/NBD
test MBG/CNBD-k
test Pareto/NBD (HB) Ma/Liu
drop unused C++ methods: post_gamma, slice_sample_gamma, post_mvnorm, slice_sample_mvnorm
etc...

could not find function "plotSampledTimingPatterns"

On a fresh BTYDPlus install I am not able to run demo("mbg-cnbd-k"). I get the following error message:

' Plot Timing Patterns of a few sampled customers

plotSampledTimingPatterns(groceryElog, T.cal = "2006-12-31")
Error in eval(expr, envir, enclos) :
could not find function "plotSampledTimingPatterns"

FormatR or Lintr for Code reformatting

http://romantsegelskyi.github.io/blog/2015/11/16/good-practices-r-package/

add extra safety to dc.check.model.params

check whether names of passed numeric vector match printnames

allow date.zero to be set in *.GenerateData methods

Rcpp warning

warning: suggest parentheses around comparison in operand of '&' [-Wparentheses]
if (add < 1e-8 & i >= 100) {

add grocery dataset for demo purposes

slice_sample_cpp loop does not finish

structure(list(cust = c(103523979L, 106539747L, 109669649L, 114536229L, 
115893543L, 115959927L, 117271474L, 121815850L, 123091742L, 123545940L, 
126225453L, 126396424L, 126844622L, 126963045L, 132218991L, 133320753L, 
134979819L, 135327695L, 136081180L, 136651599L, 137087039L, 138027869L, 
138717722L, 139202404L, 141638965L, 142495136L, 143705614L, 145281013L, 
147186313L, 147408159L, 149208624L), x = c(0, 0, 0, 0, 1, 0, 
0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 1, 1, 0, 0, 0, 0, 4, 0, 0, 0, 
1, 0, 0, 0), t.x = c(0, 0, 0, 0, 20.1428571428571, 0, 0, 0, 0, 
0, 0, 3, 0, 0, 0.285714285714286, 0, 0, 16.8571428571429, 15.4285714285714, 
0, 0, 0, 0, 9.28571428571429, 0, 0, 0, 16.1428571428571, 0, 0, 
0), litt = c(0, 0, 0, 0, 3.00284974132285, 0, 0, 0, 0, 0, 0, 
1.09861228866811, 0, 0, -1.25276296849537, 0, 0, 2.82477447541035, 
2.73622107806891, 0, 0, 0, 0, 2.06327660482648, 0, 0, 0, 2.78147766965703, 
0, 0, 0), sales = c(154.98, 849.98, 269.94, 54.98, 499.43, 249.99, 
106.98, 944.97, 36.67, 349.97, 24.99, 2149.93, 257.71, 889.95, 
37.97, 29.99, 349.99, 1494.92, 594.92, 977.93, 479.99, 59.99, 
204.96, 7566.69, 54.98, 839.97, 14.99, 2914.91, 192.98, 208.94, 
1029.92), sales.x = c(0, 0, 0, 0, 162.98, 0, 0, 0, 0, 0, 0, 869.95, 
0, 0, 29.98, 0, 0, 274.94, 149.99, 0, 0, 0, 0, 1698.89, 0, 0, 
0, 139.99, 0, 0, 0), first = structure(c(12059, 12056, 12070, 
12079, 12075, 12056, 12077, 12066, 12058, 12070, 12059, 12078, 
12062, 12054, 12055, 12054, 12055, 12071, 12069, 12056, 12077, 
12064, 12062, 12066, 12082, 12080, 12082, 12069, 12062, 12069, 
12059), class = "Date"), T.cal = c(23.8571428571429, 24.2857142857143, 
22.2857142857143, 21, 21.5714285714286, 24.2857142857143, 21.2857142857143, 
22.8571428571429, 24, 22.2857142857143, 23.8571428571429, 21.1428571428571, 
23.4285714285714, 24.5714285714286, 24.4285714285714, 24.5714285714286, 
24.4285714285714, 22.1428571428571, 22.4285714285714, 24.2857142857143, 
21.2857142857143, 23.1428571428571, 23.4285714285714, 22.8571428571429, 
20.5714285714286, 20.8571428571429, 20.5714285714286, 22.4285714285714, 
23.4285714285714, 22.4285714285714, 23.8571428571429), T.star = c(74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857, 74.7142857142857, 74.7142857142857, 
74.7142857142857, 74.7142857142857), x.star = c(0L, 0L, 0L, 1L, 
0L, 1L, 0L, 0L, 0L, 1L, 0L, 1L, 0L, 1L, 1L, 0L, 0L, 6L, 1L, 0L, 
0L, 1L, 0L, 2L, 0L, 2L, 0L, 2L, 0L, 0L, 1L), sales.star = c(0, 
0, 0, 54.9, 0, 155.97, 0, 0, 0, 299.94, 0, 379.98, 0, 210.76, 
149.99, 0, 0, 1597.82, 389.98, 0, 0, 393.92, 0, 399.97, 0, 1764.45, 
0, 1899.94, 0, 0, 49.99)), .Names = c("cust", "x", "t.x", "litt", 
"sales", "sales.x", "first", "T.cal", "T.star", "x.star", "sales.star"
), row.names = c(NA, -31L), class = "data.frame")

pggg.draws <- pggg.mcmc.DrawParameters(cbs, mcmc=2500, burnin=500, chains = 1, thin=50)

results in

Error in pggg_slice_sample("lambda", x = data$x, tx = data$t.x, Tcal = data$T.cal,  :
  slice_sample_cpp loop did not finish

Could not find mbgnbd.ConditionalExpectedTransactions

Error in mbgnbd.ConditionalExpectedTransactions(params = params.mbgnbd, :
could not find function "mbgnbd.ConditionalExpectedTransactions"

Dockerfile

More a proposal than an issue: I wrote a dockerfile for BTYDplus, any interest to integrate it in the project?

Why are hyper_priors all 0.001?

Should these be optional input parameters? do they have a strong effect if changed?

add pnbd.GenerateData

which wraps pggg.GenerateData

auto-guess `T.tot` in `bgcnbd.PlotTrackingInc`

and allow for skipping the inc argument, as well as allow for longer time horizons to be plotted

cdnow <- cdnow.sample()
elog  <- cdnow$elog
cbs   <- cdnow$cbs
(params <- bgcnbd.EstimateParameters(cbs))
inc <- elog2inc(elog)
bgcnbd.PlotTrackingInc(params, cbs$T.cal, T.tot = 78, inc) # check e.g. T.tot = 100

also, reconsider the logic for by=7 in elog2cum

interactive online demo of BTYDplus

Let's supplement the R package with a Shiny app to demo the provided BTYDplus functionality (see this tutorial). As a second step, the Shiny App could then be hosted at www.shinyapps.io or at our own Shiny Server.

UI Flow:

upload transaction data (with cust and date fields) and specify calibration period cutoff
or generate random transaction data following a model's assumption and specified parameters - the parameters should have reasonable defaults already provided
select the model (NBD, Pareto/NBD, BG/CNBD-k, MBG/CNBD-k, Pareto/GGG)
estimate parameters
estimate conditional expected transactions
plot aggregate level (plotTrackingInc, plotTrackingCum)
report MAE error
report frequency distribution for holdout

Increase Code Coverage

add methods to MCMC models

we should have:

mcmc.ExpectedCumulativeTransactions
mcmc.Expectation
mcmc.PlotTrackingCum
mcmc.PlotTrackingInc
mcmc.PlotFrequencyInCalibration

accept tibbles as input

e.g. for elog2cbs (in addition to data.frame and data.table)

speedup mcmc.pmf

Extract individual level drwas from draws$level_1

Provide helper method which merges MCMC chains

data.table.rdb is corrupt error on installing BTYDplus

I've run into this problem before, and recall solving it, but am at a loss today.

I get this error when loading BTYDplus (library(BTYDplus))

Error: package or namespace load failed for 'data.table' in get(Info[i, 1], envir = env): lazy-load database '/home/mark/R/x86_64-pc-linux-gnu-library/3.4/data.table/R/data.table.rdb' is corrupt In addition: Warning message: In get(Info[i, 1], envir = env) : internal error -3 in R_decompress1

I've tried removing data.table and BTYDplus, and reinstalling, but no luck.

I've installed with devtools::install_github("mplatzer/BTYDplus", dependencies=TRUE)

benchmark predictive accuracy of Pareto/NBD Abe when adding covariates

add example for estimating monetary component

> setDT(elog)
> spends <- elog[, .(m.x=mean(sales), x=.N), by=cust]
> spend.params <- spend.EstimateParameters(spends[m.x>0, m.x], spends[m.x>0, x])
> cbs$avg.spend <- spends$m.x
> cbs$avg.spend.star <- spend.expected.value(spend.params, spends$m.x, spends$x)
> cbs
       cust x       t.x     litt    T.cal T.star x.star avg.spend avg.spend.star
   1: 10002 0  0.000000 0.000000 33.71429     39      0  23.72333       24.70481
   2: 10020 0  0.000000 0.000000 33.71429     39      0  37.55500       36.74995
   3: 10041 0  0.000000 0.000000 33.71429     39      0   6.79000       13.84195

elog2cbs should handle strings for T.cal and T.tot

Add Vignette

http://r-pkgs.had.co.nz/vignettes.html

allow multiple T.star in pggg.generateData, pnbd.generateData and abe.generateData

so that we are consistent with bgcnbdk.generateData

enable code coverage reports

see https://github.com/jimhester/covr

bgcnbd.PlotTrackingCum / mbgcnbd.PlotTrackingCum

Just downloaded the latest version of package via command:
devtools::install_github("mplatzer/BTYDplus", dependencies=TRUE)

While I am trying to call subj functions I got an error:
Error in .Call("_BTYDplus_xbgcnbd_exp_cpp", PACKAGE = "BTYDplus", params, :
"_BTYDplus_xbgcnbd_exp_cpp" not available for .Call() for package "BTYDplus".

It seems like this function is not public anymore.

Add "m.x" to elog2cbs() output

Calculation of the average spend per repeat transaction in the calibration period ("m.x.") cannot be derived from the cbs created by elog2cbs() because "sales" appears to be a vector of TOTAL sales (ie. first transaction is NOT removed).

It would be great to provide "sales.x" (repeat sales only, first transaction removed) so one can calculate m.x = sales.x/x; and/or m.x.

allow for flexible number of draws in `mcmc.DrawFutureTransactions`

add plot.TimingSamples

TODO

add clumpy measure
add plot_samples, plot_totals,..
handling large 1mio+ customers? check out pnbd.compress.cbs
extract individual-level draws from draws$level_1?

Python version

Hi, I wonder if there is a python version of this package.
Thanks,
Nahue

Use of table in xbgcnbd.PlotFrequencyInCalibration (and mcmc.PlotFrequencyInCalibration)

Hi!

First, thanks a lot for the package. Really useful stuff here.

I noticed what I think is a bug in xbgcnbd.PlotFrequencyInCalibration (and potentially in mcmc.PlotFrequencyInCalibration, but I didn't use it directly).
The behavior at the line
x_act <- table(x_act)
is not correct, to me, when a member of the series 0:censor is missing in the table, e.g. if our censor is at 52 and nobody made 20,35, or 45 repeat transactions.
It will not fill with 0 this frequency values and hence x_act will be smaller than x_ect and due to the way R handle this, the matrix will be ... strange (got a warning in Rstudio, didn't in Visual Studio)
It can be fixed by using:
x_act <- table(factor(x_act, levels=c(0:censor)))

See this toy example to understand what i mean:

censor=25

x_act <- c(1:10, 20:30)
x_act[x_act > censor] <- censor
x_act <- table(x_act)
x_est <- 0:censor
mat <- matrix(c(x_act, x_est), nrow = 2, ncol = censor + 1, byrow = TRUE)
mat

x_act2 <- c(1:10, 20:30)
x_act2[x_act2 > censor] <- censor
x_act2 <- table(factor(x_act2,levels=c(0:censor)))
x_est <- 0:censor
mat2 <- matrix(c(x_act2, x_est), nrow = 2, ncol = censor + 1, byrow = TRUE)
mat2

I know this is a limit case. But I encountered it analysing some 'small' data subset and thought it would be worth rising it up here.
I'd gladly submit a pull request if you want.

Cheers,
Sébastien

Ps: first post actually on github... please don't hate me if I'm too long...

add tests for recoverability of Pareto/NBD (Abe) parameters with custom covariates

mbgcnbd.PlotTrackingInc legend parameter

add possibility to change legend parameter

Warn if bgnbd.ConditionalExpectations will take long

For many distinct values of T.star the computation will take long. Print out warning plus advice how to speed up

avoid returning of data.tables

the R community is not necessarily familiar with data.table, so let's ensure that methods always return data.frames, and that we don't use data.table syntax in demo/help code

can't use BTYDplus package(c++ function)

param.draws <- pnbd.mcmc.DrawParameters(data,mcmc = 20000, burnin = 3000, thin = 20, chains = 1, param_init =list(r=1,alpha=0.5,s=0.5,beta=1)) # short MCMC to run demo fast

Error in .Call("_BTYDplus_slice_sample_gamma_parameters", PACKAGE = "BTYDplus", :
"_BTYDplus_slice_sample_gamma_parameters" not available for .Call() for package "BTYDplus"

mcmc.PlotTrackingCum throws error for Pareto/NBD (Abe)

(bug report thanks to Keith Botner)

cdnowElog <- read.csv(system.file("data/cdnowElog.csv", package = "BTYD"),
                      stringsAsFactors = FALSE,
                      col.names = c("cust", "sampleid", "date", "cds", "sales"))

cdnowElog$date <- as.Date(as.character(cdnowElog$date), format = "%Y%m%d")
cbs <- elog2cbs(cdnowElog, T.cal = "1997-09-30", T.tot = "1998-06-30")

first <- aggregate(sales ~ cust, cdnowElog, function(x) x[1] * 10^-3)
names(first) <- c("cust", "first.sales")

cbs <- merge(cbs, first, by = "cust")

draws.m2 <- abe.mcmc.DrawParameters(cbs,
                                    covariates = c("first.sales"),
                                    mcmc = 7500, burnin = 2500)
weekly_cum_repeat <- elog2cum(cdnowElog)
cum_vs_exp_repeat <- mcmc.PlotTrackingCum(draws.m2,
                                          T.cal = cbs$T.cal,
                                          T.tot = round(max(cbs$T.cal + cbs$T.star)),
                                          weekly_cum_repeat,
                                          sample_size = 10000)

throws

Error in covars %*% params$beta : 
  requires numeric/complex matrix/vector arguments

fix ggnbd.ConditionalExpectedTransactions

mbg-cnbd-k: ConditionalExpectedTransactions - why length >= 100

this may be a dumb one... The model mbg-cnbd-k.R has the ConditionalExpectedTransactions function, where there is code to correct for bias. My question is: why length >= 100? is this some rule of thumb? what's the rational behind it? This makes the forecasts conditional to the length of the vector, and not only to their past behaviour. Am I missing something?

I also do not understand the comment: "Only do so, if we can safely assume that the full customer cohort is passed."

# Adjust bias BG/NBD-based approximation by scaling via the Unconditional
# Expectations (for wich we have exact expression). Only do so, if we can
# safely assume that the full customer cohort is passed.
do.bias.corr <- k > 1 && length(x) == length(t.x) 
                            && length(x) == length(T.cal) && length(x) >= 100

PGGG: lower upper bound limit of k slice sampling

When drawing pggg parameters I tend to get a limited (0.5 %) cluster of cases with mean k's between 985 and 999, with absolutely no mean k between 50 and 985 (Expected aggregate k is around 0.8).

The only thing seemingly setting these cases apart is that they're somewhat regular but rather short-lived, they're probably somewhat overrepresented and must be confusing the algorithm.

The upper bound for k slice sampling is set at 1000, which at first sight I don't think is a realistic expectation in any scenario? I would suspect a limit of around 100 to be safer and would adjust the algorithm accordingly.

Speaking of assumptions, it occurred to me that k's aggregate distribution is more likely to follow a lognormal than a gamma distribution. Even in the clumpiest of scenarios, extremely low k's remain less likely than values around 0.5, with a few higher k cases always remaining quite likely, a situation the gamma distribution doesn't allow for.

add missing BTYD plots

pnbd.PlotFreqVsConditionalExpectedFrequency
pnbd.PlotRecVsConditionalExpectedFrequency

Error when running Pareto NBD

When I'm running Pareto NBD, it threw an error:

customer_by_sufficient_statistic_pnbd <-
  transaction_data %>%
  elog2cbs(elog = ., units = "days")

params_pnbd <- BTYD::pnbd.EstimateParameters(customer_by_sufficient_statistic_pnbd)
names(params_pnbd) <- c("r", "alpha", "s", "beta")
round(params_pnbd, 3)

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

My data is transaction data sample:

knitr::kable(head(transaction_data, 50))

|cust          |date                | sales|
|:-------------|:-------------------|-----:|
|000000000000o |2016-08-24 06:36:28 |  3000|
|00000025      |2016-08-02 10:14:31 |  1000|
|00000027      |2016-08-02 13:18:19 |  3000|
|00000030      |2016-08-23 19:48:21 |  3000|
|00000030      |2016-08-25 11:25:29 |  3000|
|00000030      |2016-08-26 00:18:09 |  3000|
|00000030      |2016-08-26 13:56:38 |  1000|
|00000030      |2016-08-26 14:06:14 |  1000|
|00000112      |2016-08-05 16:32:23 |  1000|
|00000113      |2016-08-04 21:32:20 |  1000|
|00000134      |2016-08-03 17:45:27 |  1000|
|00000134      |2016-08-03 18:19:24 |  1000|
|00000134      |2016-08-04 11:10:53 |  1000|
|00000134      |2016-08-04 21:26:10 |  1000|
|00000139      |2016-08-07 00:40:19 |  1000|
|00000139      |2016-08-07 18:40:35 |  1000|
|00000139      |2016-08-07 18:41:35 |  1000|
|00000139      |2016-08-09 14:54:24 |  1000|
|00000139      |2016-08-09 14:55:31 |  1000|
|00000148      |2016-08-04 17:47:40 |  3000|
|00000148      |2016-08-04 18:36:44 |  5000|
|00000148      |2016-08-05 01:43:27 |  5000|
|00000148      |2016-08-05 11:53:36 |  1000|
|00000148      |2016-08-05 14:34:41 |  1000|
|00000148      |2016-08-10 13:15:12 |  5000|
|00000148      |2016-08-11 11:02:51 |  5000|
|00000148      |2016-08-12 13:38:01 |  5000|
|00000179      |2016-09-25 13:21:36 |  1000|
|00000180      |2016-08-02 19:16:07 |  3000|
|00000188      |2016-08-04 06:24:25 |  3000|
|00000188      |2016-08-04 06:24:49 |  3000|
|00000191      |2016-08-05 13:48:37 |  1000|
|00000191      |2016-08-08 12:26:30 |  1000|
|00000191      |2016-08-08 13:39:25 |  1000|
|00000191      |2016-08-08 13:41:20 |  1000|
|00000191      |2016-08-09 09:57:15 |  1000|
|00000191      |2016-08-09 13:58:22 |  1000|
|00000191      |2016-09-27 17:50:06 |  1000|
|00000222      |2016-08-02 20:09:57 |  1000|
|00000222      |2016-08-03 09:09:58 |   200|
|00000222      |2016-08-03 09:10:17 |   200|
|00000222      |2016-08-03 09:10:37 |   200|
|00000222      |2016-08-03 15:07:04 |   200|
|00000222      |2016-08-04 11:04:57 |   200|
|00000222      |2016-08-04 11:05:15 |   200|
|00000222      |2016-08-04 11:49:45 |   200|
|00000222      |2016-08-05 08:52:40 |   200|
|00000222      |2016-08-05 08:53:29 |   200|
|00000222      |2016-08-07 14:22:04 |   500|
|00000222      |2016-08-07 14:22:37 |   200|

My cbs sample:

knitr::kable(head(customer_by_sufficient_statistic_pnbd, 50))

|cust          |  x|        t.x|        litt| sales|first               |    T.cal|
|:-------------|--:|----------:|-----------:|-----:|:-------------------|--------:|
|000000000000o |  0|  0.0000000|    0.000000|  3000|2016-08-24 06:36:28 | 75.72418|
|00000025      |  0|  0.0000000|    0.000000|  1000|2016-08-02 10:14:31 | 97.57275|
|00000027      |  0|  0.0000000|    0.000000|  3000|2016-08-02 13:18:19 | 97.44512|
|00000030      |  4|  2.7624190|   -5.696879| 11000|2016-08-23 19:48:21 | 76.17426|
|00000112      |  0|  0.0000000|    0.000000|  1000|2016-08-05 16:32:23 | 94.31035|
|00000113      |  0|  0.0000000|    0.000000|  1000|2016-08-04 21:32:20 | 95.10205|
|00000134      |  3|  1.1532755|   -4.951050|  4000|2016-08-03 17:45:27 | 96.25961|
|00000139      |  4|  2.5938889|  -14.110905|  5000|2016-08-07 00:40:19 | 92.97150|
|00000148      |  7|  7.8266319|   -6.040406| 30000|2016-08-04 17:47:40 | 95.25807|
|00000179      |  0|  0.0000000|    0.000000|  1000|2016-09-25 13:21:36 | 43.44284|
|00000180      |  0|  0.0000000|    0.000000|  3000|2016-08-02 19:16:07 | 97.19664|
|00000188      |  1|  0.0002778|   -8.188689|  6000|2016-08-04 06:24:25 | 95.73255|
|00000191      |  6| 53.1676968|   -6.586634|  7000|2016-08-05 13:48:37 | 94.42407|
|00000222      | 13|  5.8879861|  -47.090180|  4100|2016-08-02 20:09:57 | 97.15926|
|00000227      |  0|  0.0000000|    0.000000|  1000|2016-08-03 14:14:16 | 96.40626|
|00000233      | 11| 67.8287731|  -16.401807| 16000|2016-08-12 22:34:53 | 87.05861|
|00000245      |  3|  2.3961111|   -1.882027|  4000|2016-08-03 09:16:11 | 96.61326|
|00000247      |  0|  0.0000000|    0.000000|  1000|2016-08-09 20:10:06 | 90.15916|
|00000252      | 13|  7.7163773|  -33.256233|  7300|2016-09-22 18:33:27 | 46.22627|
|00000268      |  1|  2.2652315|    0.817677|  4000|2016-08-02 17:04:13 | 97.28824|
|00000293      |  4|  6.7340046|   -9.404809|  5000|2016-08-02 18:00:45 | 97.24898|
|00000297      |  6|  0.9672569|  -12.129457|  7000|2016-08-02 15:58:47 | 97.33368|
|00000306      | 34| 39.8468981|  -49.574711| 62000|2016-08-02 18:23:14 | 97.23337|
|00000322      | 45| 37.1828356|  -71.839988| 46000|2016-08-05 10:16:48 | 94.57117|
|00000348      |  0|  0.0000000|    0.000000|  1000|2016-08-02 16:21:03 | 97.31822|
|00000358      |  0|  0.0000000|    0.000000|  1000|2016-08-05 14:46:34 | 94.38383|
|00000384      |  0|  0.0000000|    0.000000|  5000|2016-08-07 19:31:40 | 92.18584|
|00000388      |  0|  0.0000000|    0.000000|  1000|2016-08-02 17:38:09 | 97.26468|
|00000410      | 19|  2.9094792|  -55.245497| 24000|2016-08-02 14:25:06 | 97.39874|
|00000429      |  0|  0.0000000|    0.000000|  1000|2016-08-04 18:58:38 | 95.20878|
|00000449      | 37|  4.2021528| -188.924434| 80000|2016-08-02 13:03:11 | 97.45562|
|00000474      |  2|  0.7993750|   -2.135192|  3000|2016-08-02 20:04:38 | 97.16295|
|00000487      |  0|  0.0000000|    0.000000|  3000|2016-08-02 13:36:12 | 97.43270|
|00000508      | 19| 53.3256597|  -39.142569| 24000|2016-08-02 13:57:38 | 97.41781|
|00000533      |  1|  0.1032870|   -2.270243|  2000|2016-08-05 10:00:43 | 94.58234|
|00000543      |  1|  0.0118981|   -4.431373|  8000|2016-08-03 16:18:37 | 96.31991|
|00000550      |  0|  0.0000000|    0.000000|  5000|2016-08-02 18:17:33 | 97.23731|
|00000559      | 31| 36.2829051| -110.370241| 33900|2016-08-03 17:21:53 | 96.27597|
|00000562      |  4|  0.2167361|  -13.778951|  9000|2016-08-02 20:40:43 | 97.13789|
|00000567      |  0|  0.0000000|    0.000000|  1000|2016-08-02 19:19:55 | 97.19400|
|00000577      |  0|  0.0000000|    0.000000|  1000|2016-08-02 21:11:37 | 97.11644|
|00000580      | 30| 96.1238889|   17.855325| 31000|2016-08-03 16:11:21 | 96.32495|
|00000596      |  2|  3.9631134|    1.355179|  3000|2016-08-03 21:06:09 | 96.12023|
|0000060009    |  0|  0.0000000|    0.000000|  1000|2016-08-27 11:29:07 | 72.52095|
|00000603      |  0|  0.0000000|    0.000000|  1000|2016-08-02 22:51:40 | 97.04696|
|00000619      |  0|  0.0000000|    0.000000|  3000|2016-08-05 07:20:43 | 94.69345|
|00000637      |  2|  0.0893750|   -9.179702|  3000|2016-08-02 16:35:39 | 97.30808|
|00000654      |  0|  0.0000000|    0.000000|  3000|2016-09-15 17:13:16 | 53.28196|
|00000657      |  2| 27.8792130|    3.926673|  3000|2016-08-03 13:54:12 | 96.42020|
|00000661      |  1|  0.2140509|   -1.541541|  2000|2016-08-02 15:08:05 | 97.36889|

Does the data require any constraint in order to apply the method properly?

Thank you in advance.

Throw error for empty data frame?

I recently encountered the following error message, Error:_ T.tot >= min(elog$date) is not TRUE, which led me to investigate whether my dates were correct inside of a list of data frames. Finally, I made the connection that some of my data frames were empty, which prompted the message.

It might be worth adding a check for whether the supplied data frame is empty, and give a more informative error message. Do you agree? If so, I'll file a pull request.

library(BTYDplus)
df <- data.frame(cust = character(), date = as.Date(character()))
elog2cbs(df)

  n <- 5000
  params <- list(r = 0.9, alpha = 10, s = 0.8, beta = 12)
  # constant T.cal
  T.cal <- rep(52, n)
  cbs1 <- pnbd.GenerateData(n, T.cal, 52, params, TRUE)$cbs
  # varying T.cal, with one T.cal being particularly large
  T.cal[n] <- 52 * 100
  cbs2 <- pnbd.GenerateData(n, T.cal, 52, params, TRUE)$cbs
  expect_equal(sum(cbs1[-n]$x.star), sum(cbs2[-n]$x.star), tolerance = 0.2)