Time-Slice CV Class

[timbo112711]

Currently to implement Time-Slice CV one would need to reference the example notebook and copy / pasta the code blocks. We could wrap the code inside the notebook into a TimeSliceCV class. Then in the example notebook, we reference the needed methods to:

1. Set up Time-Slice CV
2. Run the CV method
3. Create various plotting methods for diagnostics

[juanitorduz]

yes!

[williambdean]

Is the goal to make the notebook logic easy accessible?

[timbo112711]

Is the goal to make the notebook logic easy accessible?

@williambdean, the goal would be two fold. Make the notebook logic more accessible as well as make the application easier. For example, the TimeSliceCrossValidationResult dataclass, time_slice_cross_validation_step method, and the for loop can be more streamlined. That way the users don't need to copy and paste those blocks into their workflow. All they need to do is call

from pymc_marketing.mmm import TimeSliceCV

TimeSliceCV(...)

This could reduce workflow errors and make things easier in deployment cases.

[williambdean]

Would be good to make it work with the scikit-learn / generator approach for returning the spilts

for example : https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html