New spends forward pass

[wd60622]

Current code addressing #451. Have a look @carlosagostini

I've split it into two methods:

• generation the data for the plot
• plotting of the data

I'd like to hear feedback on the arguments for the generation. Specifically, the spends and the one_time
Currently, it takes a np.ndarray that would be a raw spend for each channel (error if it doesn't match) and will ultimately transform it to a (max_lag, n_channels) ndarray with values depending on the one_time argument.
One time:

• True -> spends are the first and all rest are zeros in the (max_lag, n_channels)
• False -> spends for each channel are for each of the
  This locks into a bit of assumption of how this would be used. Maybe, the spends could be (n_channels, ) or (max_lags, n_channels) to reduce the need of one_time and simplify the method.
  Would like to hear how you might use it.

The method returns the untransformed spends although named _contributions. The inverse transformation only happen in the plotting method, but it maybe good to have the inverse transformations happening that first method.
I think this is a good case to remove the target transformation from the sklearn pipeline so it can be part of the forward pass of contributions @ricardoV94

EDIT: The inverse transformation is done in the new_spend_contributions method as well

---

📚 Documentation preview 📚: https://pymc-marketing--456.org.readthedocs.build/en/456/

[codecov]

Codecov Report

Attention: Patch coverage is 91.80328% with 5 lines in your changes are missing coverage. Please review.

Project coverage is 91.09%. Comparing base (2f2d814) to head (a523f00).
Report is 41 commits behind head on main.

Files	Patch %	Lines
pymc_marketing/mmm/delayed_saturated_mmm.py	91.11%	4 Missing ⚠️
pymc_marketing/mmm/utils.py	90.90%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #456      +/-   ##
==========================================
- Coverage   91.24%   91.09%   -0.15%     
==========================================
  Files          21       21              
  Lines        2044     2056      +12     
==========================================
+ Hits         1865     1873       +8     
- Misses        179      183       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[wd60622]

Got the prior functionality working

The inverse_transform could be used if the model was fit. Might be good to toggle inverse_transform even for the posterior. Any thoughts here?

[wd60622]

Will add the transformation back to original space and continue with this PR after #482

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[juanitorduz]

This looks very nice! I added some minor suggestions on the docstrings to clarify this better for the end-user. WDYT?

[wd60622]

I've addressed all the feedback, @juanitorduz
Thanks!

I've separated out the "new_spend" creation into a function and added some more to the docstrings so hopefully that brings in some clarity.

Let me know what you think!

[juanitorduz]

@wd60622 looks great! There is one test failing, though. If you fix it, we can merge this one :)

[wd60622]

Have the test passing locally but not in the action 😢

Running the action with 3.11 with the latest pymc to see if the pymc version has an impact. The test depends on the sample_prior_predictive method. Not sure if that has change

Might just mock the idata for all of these tests with the plots and hope giving a seed will keep the results the same

[wd60622]

Trying to mock the prior in the model.idata
Hope that that works. If not, I will pull out the plotting into a function and tests on some fixed data.

If all else fails, these are tests against the plotting images and can maybe be omitted. Thoughts? @juanitorduz @ricardoV94

EDIT: They seem to be failing again... This tests are working locally for me. Honestly super hard to check what's going on on the test comparison. Any idea if tests files are stored off? i.e. log here
Thought I'd give this pytest-mpl a try though. Inspired by this #355

[juanitorduz]

@wd60622 I am getting the same error as in matplotlib/pytest-mpl#69 . Can you please add markers =["mpl_image_compare"] to pyproject.toml as in

[tool.pytest.ini_options]
addopts = [
    "-v",
    "--strict-markers",
    "--strict-config",
    "--cov=pymc_marketing",
    "--cov-report=term-missing",
    "--color=yes",
]
markers =["mpl_image_compare"]
filterwarnings = ["ignore::DeprecationWarning:bokeh.core.property.primitive:37"]
testpaths = "tests"

[wd60622]

Failed again even after using fresh environment 😢
Would you be able to check one more time @juanitorduz

[juanitorduz]

Failed again even after using fresh environment 😢 Would you be able to check one more time @juanitorduz

I suggest we make sure the plot runs (as other plots tests) and create an issue to add this matplotlib testing framework. I would not delay this feature because of this :)

[wd60622]

Failed again even after using fresh environment 😢 Would you be able to check one more time @juanitorduz

I suggest we make sure the plot runs (as other plots tests) and create an issue to add this matplotlib testing framework. I would not delay this feature because of this :)

Sounds great!

[juanitorduz]

Thanks @wd60622 !

[wd60622]

Thanks @wd60622 !

Thank you for the review!