1.2.0 - add `legend` argument to `(m)bgcnbd.PlotTrackingInc`, `(m)bgcnbd.PlotTrackingCum`, `mcmc.PlotTrackingCum`, and `mcmc.PlotTrackingInc` - made vignette code compatible with BTYD 2.4.2 - deprecated inefficient alternative sampling method for Pareto/NBD MCMC - updated referenced URLs (to HTTPs or redirected pages) 1.1.4 - `elog2cbs` handles now empty elogs gracefully (#58) 1.1.3 - limit simulated life length to 10k events in `xbgcnbd.GenerateData` 1.1.2 - fix `mcmc.ExpectedCumulativeTransactions` for Pareto/NBD (Abe) with covariates (#55) - add ability to provide covariates to `abe.generateData` 1.1.1 - added `sales.x` to `elog2cbs` output, which sums over sales in calibration, but excludes initial transaction (thanks to msinjin for the suggestion) 1.1.0 - added `(m)bgcnbd.PlotFreqVsConditionalExpectedFrequency` methods - added `(m)bgcnbd.PlotRecVsConditionalExpectedFrequency` methods 1.0.2 - speed up `xbgcnbd.EstimateParameters` by lowering lower bound of early stop to k=3 1.0.1 - reduce run time of tests and example code - add source info to `groceryElog` dataset 1.0.0 - updated package Vignette - bump in major version number to reflect intended first CRAN release 0.11.x - bug fix in Pareto/GGG demo 0.11.3 - bug fix in MBG/CNBD-k demo which used a deprecated method name 0.11.2 - add option `date.zero` to specify start date for cohort in `*.GenerateData` methods 0.11.1 - hot fix for broken `elog2cbs` in `0.11.0` release 0.11.0 - rewrote `*.GenerateData` methods * generated event log now also contains `date` field * if `T.cal` varies, then customer's start date will also vary * use `elog2cbs` internally to convert generated event log to CBS format; this also adds `first` to CBS * `*.GenerateData` now always returns the event log -> argument `return.elog` has been removed * fix bug for `*.GenerateData` in case of with highly varying `T.cal` (fixes #42) - fix top margin of `mcmc.plotPActiveDiagnostic` 0.10.2 - avoid internal C++ compile warning message - R CMD CHECK now returns 0 errors and 0 warnings 0.10.1 - add lintr checks for adhering to consistent code style - run spellcheck for docs 0.10.0 - add a package vignette - add AppVeyor for automated build tests on Windows - rename `plotSampledTimingPatterns` to `plotTimingPatterns` - rename argument `per` to `units` for `elog2cbs` to be consistent with `difftime` - fix `elog2cbs`: allow T.tot to be larger than max(elog$date) - fix `elog2inc` and `elog2cum` to not report last week, if that is not complete - only do bias correction for `mbgcnbd.ConditionalExpectedTransactions` when we can safely assume that the whole customer cohort is passed - `abe.mcmc.DrawParameters` can now handle data.table's - add ability to determine minimum number of transactions per customers to `estimateRegularity` 0.9.0 - add argument `first` to `elog2cum` and `elog2inc` - bugfix for `elog2inc` and `elog2cum` for elogs where dates are missing - add `groceryElog` dataset, and remove `cdnow.sample()` - use `groceryElog` dataset for demos and examples - `elog2cbs` prunes columns, if `elog$sales` or `T.cal` is not provided 0.8.3 - rewrite demos - add `plotSampledTimingPatterns` for plotting samples of timing patterns - add `mbgnbd.EstimateParameters` to estimate MBG/NBD model 0.8.2 - improve documentation - remove MBG/NBD methods, as these are easily accessible via the MBG/CNBD-k methods - allow `T.star` to be a vector in `*.GenerateData` methods for MCMC models 0.8.0 - add * `pnbd.GenerateData` * `mcmc.pmf` * `mcmc.Expectation` * `mcmc.mcmc.ExpectedCumulativeTransactions` * `mcmc.PlotTrackingCum` * `mcmc.PlotTrackingInc` * `mcmc.PlotFrequencyInCalibration` - adapt `bgcnbd.pmf` to return matrix in case that both `t` and `x` are vectors - dropp unnecessary `cal.cbs` argument from `mcmc.PAlive` - allow flexible sample size in `mcmc.DrawFutureTransactions` - removeGammaGompertz/NBD due to broken `ggnbd.ConditionalExpectedTransactions` 0.7.2 - add new methods: `elog2inc`, `elog2cum`, `cdnow.sample` - improve documentation and demos 0.7.1 - `elog2cbs` now returns sum over sales for calibration and holdout period - enhance CDNow demo with estimating of monetary component (i.e. CLV estimate) - methods accepting `params` check for names, if named vectors are passed 0.7.0 - disable `ggnbd.ConditionalExpectedTransactions` - (m)bgcnbd.ConditionalExpectedTransactions uses same h2f1 method as BTYD, if GSL is not present - add Travis-CI - format R code with `formatR::tidy_dir('R', indent=2)` 0.6.4 - (m)bgcnbd.GenerateData can now handle multiple holdout period lengths, and differing calibration periods for each customer - fix bias in (m)bgcnbd.ConditionalExpectedTransactions by re-scaling via (m)bgcnbd.Expectation 0.6.3 - unconditional expectations for (M)BG/CNBD-k (`(m)bgcnbd.Expectation`) are now calculated `exact` by utilizing `(m)bgcnbd.pmf` 0.6.1 - add implementation of: (m)bgcnbd.Expectation (m)bgcnbd.PlotFrequencyInCalibration, (m)bgcnbd.ExpectedCumulativeTransactions, (m)bgcnbd.PlotTrackingCum, (m)bgcnbd.PlotTrackingInc 0.6.0 - add implementation of BG/CNBD-k model - (m)bgcnbd.ConditionalExpectedTransactions uses original approximation again - mbgcnbd.ConditionalExpectedTransactions can now handle customers with many (170+) transactions; thanks to Andrea Rumenjak for providing the patch; 0.5.0 - rename model from CBG/CNBD to MBG/CNBD - function prefixes changed from cbgcnbd to mbgcnbd accordingly - improve approximation logic for mbgcnbd.ConditionalExpectedTransactions 0.4.0 - rename model from Pareto/CNBD to Pareto/GGG - function prefixes changed from pcnbd to pggg accordingly - cust-column is assigned as names to list of level_1 CODA-objects returned by MCMC drawParameter methods - this makes it easier to access these via customer ID - rename to cbgcnbd.EstimateRegularity() to estimateRegularity(), because the method is not bound to CBG/CNBD-k model - add tests and demo 0.3.3 - add mc.cores parameter to MCMC methods to explicitely set number of parallel cores to be used - important bug-fix for pcnbd.mcmc.DrawParameters - draw_tau slice-samples now from correct log_posterior in churn case - more realistic test case for P/NBD MCMC, plus minor change to handling numeric underflow resulting in more accurate parameter estimates - pcnbd.GenerateData now correctly considers fixed parameter value for k - pcnbd.GenerateData has minimum level of 0.1 for generated k's in order to avoid itt's if 0 0.3.2 - add mcmc.setBurnin method which cuts away (additional) initial burnin steps of MCMC chains - `*.EstimateParameters` and `*.mcmc.DrawParameters` now also accept data.tables as input - pcnbd.mcmc.plotRegularityRateHeterogeneity: add option to plot point estimates - add mcmc.PActive and mcmc.plotPActiveDiagnostic 0.3.1 - deprecated `abe.mcmc.DrawFutureTransactions` and `pcnbd.mcmc.DrawFutureTransactions` in favor of newly add `mcmc.DrawFutureTransactions` method - deprecated `abe.mcmc.PAlive` and `pcnbd.mcmc.PAlive` in favor of newly added `mcmc.PAlive` method - `*.mcmc.DrawParameters` now all consistenly return sampled `z` values in their output