Note : links below send to the Novecare (CS) project, but equivalent objects exist in the SpP one also.
Refer to this article for the branch management part, including creation, rebase and merge of the branch.
Once the branch is created :
- Launch a first run up to the weights model on its own version (update "versioning_version_name" variable).
- Prepare and merge the new column in the "Data preparation" Flow zone.
- Make sure to include it in the right category in the variables (numerical_features, categorical_features, ordinal_features).
- Filter out values using the "preprocessing_filters" variable if needed.
- Impute values using the "imputers" variable if needed.
- Check the "Correlation_analysis" notebook to make sure the new feature is not highly correlated with another one.
- Launch a second run up to the weights model on its own version (update "versioning_version_name" variable).
- Check the models R2.
- Use the "SHAP version comparison" notebook to compare the SHAP values between the two runs.
- If the feature seem important and the business agrees to keep it, finish the run (until final output) and provide results for validation.
- Check that the volume curves are still fine.
- If this development is prior to a release, check group volume adjustment notebook and update the list of families to use it on on the variables.
- If this development is prior to a release, conduct hyperparameter optimization.
- Rebase and merge the branch back to dev and rerun entirely there to keep the history of the version.