Patrik Schilter
Patrik Schilter
Opening PR to run tests
## Background `d` in the pnbd dyncov model is the number of periods until the end of the covariate interval containing a transaction ('ceiling_date' on the time unit). Its definition...
Currently, the in-sample unconditional expectation is plotted in the holdout period. The summed CET would be a more appropriate choice for the holdout period. Add this with an option to...
Provide estimates of the variance of CET for pnbd as described in [Fader and Hardie (2005)](https://pdfs.semanticscholar.org/c971/3415a2faf8258e94d1052588b5d8d3a12902.pdf), page 10-11. Intervals could be reported in the prediction table together with the CET...
While introducing tests against the dyncov LL in #110 it was found that not all tests as described in #98 hold. `bkT=T*C` is not true for gamma=0: Instead of 38...
The actual and expected repeat transactions are shifted by a period because they currently are (wrongfully) defined differently: Repeat transactions are currently implemented as "forward-looking", ie the number of transactions...
While implementing the BGBB model, it became clear that the transaction history does not suffice as input for all models. Because the BGBB model is for a discrete-time setting, it...
I'd suggest to change the standard optimization method to Nelder-Mead for all models. BFGS breaks if the LL returns NA/Inf due to numerical stability issues which seem to arise rather...
Add additional period definitions: - [ ] 14d / biweekly - [ ] 28d / month.28d - [ ] 30d / month.30d