Lesson 35 – Creating a Machine Learning Model with Microsoft Fabric

This blog guides you on maximizing Microsoft Fabric and MLflow for creating, managing, and applying machine learning models effortlessly. Explore the simplicity of handling data patterns and model versions for effective data-driven outcomes.

Machine learning model in Microsoft Fabric

In Microsoft Fabric, a machine learning model is like a knowledgeable file that learns from data patterns and predicts outcomes. With MLflow, a user-friendly tool integrated into Fabric, you can easily create, manage, and use these models. It’s like having a smart assistant for handling all things related to data and AI services in one place, from training models to comparing different versions effortlessly.

How to create a machine learning model?

You have the option to create a machine learning model directly within the Fabric UI. Alternatively, the MLflow API allows for the direct creation of the model as well.

Follow the steps to create machine learning model

• Launch https://app.powerbi.com/
• In navigation pane on the left click Create.
• Under Data science workloads click on the item “ML model“.

• Enter model name and click Create.

• Register or save experiment runs (as different versions) directly to an existing model.

Manage multiple versions within a machine learning model

Managing versions within a machine learning model is like keeping a detailed history of changes in the model over time. This practice enhances the reproducibility, collaboration, and reliability of the machine learning process. It allows for easy comparison of different versions and the ability to go back to a stable version if needed.

• Within a model, a data scientist can navigate across various model versions to explore the underlying parameters and metrics.

Each version of the model includes the following details:

• Time Created: Indicates the date and time of the model’s creation.
• Run Name: A unique identifier associated with the experiment runs contributing to the specific model version.
• Hyperparameters: Configuration settings saved as key-value pairs. Both keys and values are represented as strings.
• Metrics: Numeric run metrics stored as key-value pairs, providing valuable performance indicators.
• Model Schema/Signature: A detailed description outlining the model’s inputs and outputs, providing insights into its structure and functionality.
• Logged Files: Various files, such as images, environment details, models, and data files, are recorded. This comprehensive record allows for a thorough understanding of the model’s composition and the context in which it operates.

Compare machine learning models

Data scientists can also make comparisons across model versions to identify whether or not newer models might yield better results.

• Choose an existing machine learning model with multiple versions.
• Click View tab –> Model list view.
• Customize the table columns by expanding the Customize columns pane. Select the specific properties, metrics, and hyperparameters you want to display.
• Finally, in the metrics comparison pane, select multiple versions for comparison.
• Customize charts by making changes to the chart title, visualization type, X-axis, Y-axis, and more to tailor your analysis.

• This streamlined process ensures a comprehensive comparison of different model runs for informed decision-making.

Apply machine learning models

After training a model on a dataset, you can employ it to make predictions on unfamiliar data, a process commonly known as scoring or inferencing. This step involves utilizing the model’s learned patterns to generate insights or forecasts for data it has not encountered during training.