Merge pull request #104311 from lobrien/1486482-Retrain-Models

When and how to retrain models

Author: PRMerger12

articles/machine-learning/concept-model-management-and-deployment.md

@@ -134,6 +134,19 @@ This information helps you understand how your model is being used. The collecte
 
 For more information, see [How to enable model data collection](how-to-enable-data-collection.md).
 
+## Retrain your model on new data
+
+Often, you'll want to update your model, or even retrain it from scratch, as you receive new information. Sometimes, receiving new data is an expected part of the domain. Other times, as discussed in [Detect data drift (preview) on datasets](how-to-monitor-datasets.md), model performance can degrade in the face of such things as changes to a particular sensor, natural data changes such as seasonal effects, or features shifting in their relation to other features. 
+
+There is no universal answer to "How do I know if I should retrain?" but Azure ML event and monitoring tools previously discussed are good starting points for automation. Once you have decided to retrain, you should: 
+
+- Preprocess your data using a repeatable, automated process
+- Train your new model
+- Compare the outputs of your new model to those of your old model
+- Use predefined criteria to choose whether to replace your old model 
+
+A theme of the above steps is that your retraining should be automated, not ad hoc. [Azure Machine Learning pipelines](concept-ml-pipelines.md) are a good answer for creating workflows relating to data preparation, training, validation, and deployment. Read [Retrain models with Azure Machine Learning designer (preview)](how-to-retrain-designer.md) to see how pipelines and the Azure Machine Learning designer fit into a retraining scenario. 
+
 ## Automate the ML lifecycle 
 
 You can use GitHub and Azure Pipelines to create a continuous integration process that trains a model. In a typical scenario, when a Data Scientist checks a change into the Git repo for a project, the Azure Pipeline will start a training run. The results of the run can then be inspected to see the performance characteristics of the trained model. You can also create a pipeline that deploys the model as a web service.