Add Individual Shifted Beta Geometric (sBG) model

[ricardoV94]

Closes #40
Closes #97

ToDO

• Tests
• Add more text/context to notebook
• Docstrings

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[larryshamalama]

Some minor comments and questions for now! 🙂

[ColtAllen]

I dug up a link to the research paper because I'm not familiar with this model:

https://faculty.wharton.upenn.edu/wp-content/uploads/2012/04/Fader_hardie_jim_07.pdf

Are there plans to add any of the other expressions in that paper like aggregate retention rate? There's even a discounted lifetime value function.

Also, looking over the code I see a lot of redundancy in self.model, distribution_customer_churn_time() and _distribution_new_customer(). Is there a way these could be refactored to cut down on the repeated code lines?

[ricardoV94]

I dug up a link to the research paper because I'm not familiar with this model:

https://faculty.wharton.upenn.edu/wp-content/uploads/2012/04/Fader_hardie_jim_07.pdf

Are there plans to add any of the other expressions in that paper like aggregate retention rate? There's even a discounted lifetime value function.

Yeah we can add them. In this case it's trivial to do whatever summary statistics we want since we have all the variables in the trace, including individual parameters. Once we do the model per-cohort we will need to look at the close form solutions in those papers. @ColtAllen Do you mind opening an issue for that?

Also, looking over the code I see a lot of redundancy in self.model, distribution_customer_churn_time() and _distribution_new_customer(). Is there a way these could be refactored to cut down on the repeated code lines?

Hmm... maybe, but I think it's simple enough. Just defining a new PyMC model that we can do posterior predictive on with the old fit.

[ricardoV94]

Changed t_end to T, and added missing tests. Ready for a final review

[codecov]

Codecov Report

Merging #133 (afb79f7) into main (d410c11) will increase coverage by 0.32%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##             main     #133      +/-   ##
==========================================
+ Coverage   94.12%   94.45%   +0.32%     
==========================================
  Files          17       18       +1     
  Lines         919      973      +54     
==========================================
+ Hits          865      919      +54     
  Misses         54       54              

Impacted Files	Coverage Δ
pymc_marketing/clv/__init__.py	100.00% <ø> (ø)
pymc_marketing/clv/models/__init__.py	100.00% <100.00%> (ø)
pymc_marketing/clv/models/basic.py	98.52% <100.00%> (+0.41%)	⬆️
pymc_marketing/clv/models/beta_geo.py	98.85% <100.00%> (-0.08%)	⬇️
pymc_marketing/clv/models/gamma_gamma.py	98.00% <100.00%> (-0.15%)	⬇️
pymc_marketing/clv/models/shifted_beta_geo.py	100.00% <100.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

[juanitorduz]

Closes #4

ToDO

• Tests
• Add more text/context to notebook
• Docstrings

I think the indicated issue is not correct (is pointing to MMM transformations)

[ricardoV94]

Closes #4
ToDO

• Tests
• Add more text/context to notebook
• Docstrings

I think the indicated issue is not correct (is pointing to MMM transformations)

Thanks, it used to be #40, but one of the edits killed that zero :)

[ricardoV94]

Addressed comments, ready for re-review

[juanitorduz]

I am not fully aware of the model details (I will have two), but overall the code looks nice to read. I let a small comment regarding a typo.

[juanitorduz]

I read the paper looks great! I could support adding some formulas missing from the paper in another PR! So I think if you correct the typo above we can merge this one 🙂

[ricardoV94]

Typo fixed @juanitorduz

[ricardoV94]

Found a typo in the notebook, fixing