This is a profiles application project that builds a churn prediction score on top of shopify features. The shopify features is a library project, and churn prediction comes on an independent python package that is compatible with profiles.
This project has the same requirements as the usual profiles projects. But the python_model has one extra requirement, to create a virtual environment with the required packages installed, and point to that path in the site_config. The exact steps are present in the python_model README.md
- Clone the shopify features repo and make any required changes to the entity vars. This workflow should be identical to how any regular profiles project is set up
- The github url of profiles project with feature table is added in pb_project.yaml as a package, replacing the current shopify project url. It can be a local file path as well.
- In profiles.yaml, modify the config in train/data to point to the correct feature table and target column.
- The label_column should be the entity var that needs to be predicted in advance (defined by prediction_horizon_days).
- The label_value tells the actual value in the feature table that corresponds to the users who performed the action historically.
- model_name is the name of feature table defined in the profiles project (step 1)