Integrating machine learning predictions from Azure Machine Learning (Azure ML) services in Power BI reports requires taking several measures to ensure smooth integration, data refresh compatibility, and scalability, among other factors. Here is the best approach.
Publish The Azure ML Model And Make It Available For Consumption
To begin with, it is important to publish the trained Azure machine learning model as a web service. This is accomplished by creating an endpoint, which can be done using Azure ML studio or Azure ML SDK. When the endpoint is up and running, it will offer a REST API that can be interfaced with Power BI to retrieve predictions. It is important to make sure that the API provides a proper protection mechanism, such as an API key or Active Directory Access (AAD) tokens to protect the data and the model.
Connect Power BI to the Azure ML Endpoint
In Power BI, one can integrate Power Query to invoke the Azure ML REST API. Within Power Query, leverage the ‘Web’ connector to issue an HTTP request to the specified endpoint, including the required input data (e.g., features that are needed for the prediction) in the body of the request. This API will return the predictions, which can then be transformed and loaded into the Power BI Data model. However, this may involve advanced API usage, for example, M code to script over the bindings for authentication or API.
Enhance Performance and Scalability
Since the API calls are extensively time-consuming and throughputs may be limited, one should appear to plan for efficient data refreshes. With regards to large data volumes, consider utilizing the Azure ML service for batch prediction runs on a database of records and subsequently advertising the results in a database or Data Lake such as Azure SQL Database or Azure Data Lake. Power BI can be connected straight to the database for timely refreshes and minimal dependence on the APIs. Additionally, configure your Power BI data refresh schedules to run during nonworking hours to prevent them from clashing with live production workloads.
Taking these measures, it will be possible to include machine learning forecasting in the Power BI reports and make them useful by focusing on issue resolution.
