Stewardship
Models should not be deployed without a monitoring strategy. To this end, we provide a template that addresses data drift and model drift. These steps are in addition to the data testing that is built into the feature engineering pipeline.
1 Key terms
- Ground Truth: the true answer
- Predicted or Inferred value or label: the model or the machines best guess
- Training data: a sufficient sample of the training data set used compare against the new (unseen) data
- Unseen data: data that has arrived after the model has been trained and deployed. The data in production.
2 Walk through
- Clone the ‘model seedling’ template