Media transformation sampling & plotting methods

[wd60622]

Description

This adds new methods for sampling and plotting the media transformations. They are:

• sample_prior: sample the parameter distributions for the transformation
• sample_curve: sample the curve given a set of parameters. These can be prior or posterior
• plot_curve: Plot curve HDI and n random curve samples. Uses plot_curve_hdi and plot_curve_samples methods

This creates a nice pre-modeling workflow of:

import matplotlib.pyplot as plt

saturation = ...
prior = saturation.sample_prior()
curve = saturation.sample_curve(parameters=prior)
saturation.plot_curve(curve)
plt.show()

And also model analysis

mmm = ...
mmm.fit(X, y)

# Curve can now has additional dimensions like channel
curves = mmm.saturation.sample_curve(mmm.fit_result)

Related Issue

• Closes Media transforms prior sampling & curve plotting #722
• Related to #

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--734.org.readthedocs.build/en/734/

[codecov]

Codecov Report

Attention: Patch coverage is 91.05691% with 11 lines in your changes missing coverage. Please review.

Project coverage is 93.48%. Comparing base (271966b) to head (1361413).
Report is 166 commits behind head on main.

Files with missing lines	Patch %	Lines
pymc_marketing/mmm/components/base.py	89.81%	11 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #734      +/-   ##
==========================================
- Coverage   93.60%   93.48%   -0.12%     
==========================================
  Files          27       27              
  Lines        2722     2841     +119     
==========================================
+ Hits         2548     2656     +108     
- Misses        174      185      +11     

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

[juanitorduz]

Some initial comments :)

[juanitorduz]

We usually return the figure... shall we keep that convention (no strong opinion)

[wd60622]

I am leaning toward that as well.
I am adding the support to sample the curves based on parameter distributions that could include other dimensions (i.e. channel) and think support for plotting
i.e. 3 channels in the dims would lead to figure with 3 axes each with their saturation curve.

[juanitorduz]

ok! No strong preference. Just think about it from a user perspective :)

[wd60622]

Still need to add:

• plot as grid based on the coordinates of the curves

[wd60622]

This should be good to go @juanitorduz
Let me know your thoughts

The plots adapt to the "Curve" that is used as argument. For instance, unless specified, the priors will have only chain, draw coords as dims wouldn't be known / are all the same

transformation = LogisticSaturation()

parameters= transformation.sample_prior()
curve = transformation.sample_curve(parameters)
transformation.plot_curve(curve)
plt.show()

coords = {"channel": ["A", "B", "C"]}
parameters= transformation.sample_prior(coords=coords) # This could be from posterior with additional coords for transformation
curve = transformation.sample_curve(parameters)
transformation.plot_curve(curve)
plt.show()

[juanitorduz]

Two tests are failing:

FAILED tests/mmm/test_delayed_saturated_mmm.py::test_plotting_media_transform_workflow[adstock]
FAILED tests/mmm/test_delayed_saturated_mmm.py::test_plotting_media_transform_workflow[saturation]

[wd60622]

Two tests are failing:

FAILED tests/mmm/test_delayed_saturated_mmm.py::test_plotting_media_transform_workflow[adstock]
FAILED tests/mmm/test_delayed_saturated_mmm.py::test_plotting_media_transform_workflow[saturation]

Somewhere the mmm is being modified which makes the tests sensitive to the order of tests. This should be fixed but probably not best thing to have order matter

Reference: Before and After

[juanitorduz]

@wd60622 this looks great! I just left a question :)

[juanitorduz]

We have these bad guys failing

FAILED tests/mmm/components/test_base.py::test_new_transformation_plot_curve[coords0-1]
FAILED tests/mmm/components/test_base.py::test_new_transformation_plot_curve[coords1-3]

:D

[juanitorduz]

I will update the branch and merge! Cool work!