Introduce BetaGeoNBD Random Variable

[PabloRoque]

Description

Context: Prepares the introduction of time-invariant covariates in (M)BG/NBD models

• Introduces BetaGeoNBDRV contributing to CLV API Standardization #527
• distribution_new_customer | distribution_new_customer_dropout now accept data as Optional. This is in line with other distributions, contributing to the standardization
• Adds distribution_new_customer_recency_frequency in line with standardization
• Overrides ModifiedBetaGeoModel.distribution_new_customer. Note that we cannot use super().distribution_new_customer until a new distribution block is added for ModifiedBetaGeoRV

Related Issue

• Closes #
• Related to CLV API Standardization #527 CLV Distribution RVs not Model-Specific #128
• Related to List of Continuous, Non-Contractual `clv` Models to Consider Adding #1230

Checklist

• Checked that the pre-commit linting/style checks pass
• Included tests that prove the fix is effective or that the new feature works
• Added necessary documentation (docstrings and/or example notebooks)
• If you are a pro: each commit corresponds to a relevant logical change

Modules affected

• MMM
• CLV

Type of change

• New feature / enhancement
• Bug fix
• Documentation
• Maintenance
• Other (please specify):

---

📚 Documentation preview 📚: https://pymc-marketing--1317.org.readthedocs.build/en/1317/

[ColtAllen]

It's a good start, but it looks like a lot of changes were copied over from #1269. This will create merge conflicts and/or require a rebase to preserve @ricardoV94's contributions in that PR.

Can you revert the changes to the other distribution blocks? I would prefer to merge #1269 before proceeding with this one.

[PabloRoque]

@ColtAllen Thanks for taking a look at the draft PR

It's a good start, but it looks like a lot of changes were copied over from #1269

Indeed. I need to work on top of pymc>=5.19.1 and pytensor>=2.26.1 [since it's the only way to make blas work on apple silicon without having to hack c-compiler flags]. Due to this I need to make changes to the signatures just like #1269

So I guess we will need to leave the PR as draft until #1269 is merged. Then we can rebase.

Can you revert the changes to the other distribution blocks? I would prefer to merge #1269 before proceeding with this one.

As explained above I cannot revert, because basically nothing runs on my setup if reverted.

[ColtAllen]

As explained above I cannot revert, because basically nothing runs on my setup if reverted.

Here's a workaround for MacOS.

[ColtAllen]

Looks good so far! Just need to revise the sim_data function, docstrings, and a few other things.

[ColtAllen]

Almost there!

[ColtAllen]

Thanks for adding this! This will allow use of the utils.plot_purchase_pmf function for BetaGeoModel, as well as permit arviz metrics like WAIC for out-of-sample predictive power.

If you wish to create an equivalent RV for ModifiedBetaGeoModel, you can probably just inherit from these classes and override the rng_fn and logp methods.

[PabloRoque]

I truly appreciate the time you took to go over it.
Expect that PR you mention, and one introducing covars for each of the 2 models.

[PabloRoque]

@wd60622 Could you have a look at this PR? If you don't see any issues could you consider merging it?

I am preparing ModifiedBetaGeoNBDRV and would be good to rebase to main with this PR merged.

Thanks!

[wd60622]

what was happening here?

[PabloRoque]

@pytest.mark.parametrize("model_cls", [BetaGeoModel])
    def test_clv_fit_mcmc(model_cls, clv_data) -> None:
        mlflow.set_experiment("pymc-marketing-test-suite-clv")
    
        sampler_config = {
            "draws": 2,
            "chains": 1,
            "tune": 1,
        }
    
        model = model_cls(data=clv_data, sampler_config=sampler_config)
        with mlflow.start_run() as run:
            model.fit()
    
        assert mlflow.active_run() is None
    
        run_id = run.info.run_id
        inputs, params, metrics, tags, artifacts = get_run_data(run_id)
    
>       assert inputs == []
E       assert [<DatasetInpu...e='sample'>]>] == []
E         Left contains one more item: <DatasetInput: dataset=<Dataset: digest='cebda4ee', name='dataset', profile=('{"features_shape": {}, "features_size": ... '"mlflow.source.type": "LOCAL"}}'), source_type='code'>, tags=[<InputTag: key='mlflow.data.context', value='sample'>]>
E         Full diff:
E           [
E         -  ,
E         +  <DatasetInput: dataset=<Dataset: digest='cebda4ee', name='dataset', profile=('{"features_shape": {}, "features_size": {}, "features_nbytes": {}, '
E         +  '"targets_shape": {"recency_frequency": [4, 2]}, "targets_size": '
E         +  '{"recency_frequency": 8}, "targets_nbytes": {"recency_frequency": 64}}'), schema=None, source=('...
E         
E         ...Full output truncated (4 lines hidden), use '-vv' to show

tests/test_mlflow.py:486: AssertionError

[PabloRoque]

inputs is not an empty array

[wd60622]

I see. thanks for the traceback

[wd60622]

Looks good. I just had question about the mlflow test. I'll merge still