Data Science happens on the same platform as business users and analysts. Sharing and collaboration is more seamless than before.
Fabric Spark is open. You may use Machine learning library of your choice. Model training can be tracked by MLFlow experiments. Predict function is available across the workloads and supports mlflow packaged models in the Fabric model registry.
Qumio has helped customers to create models for various business needs. Few examples are model to classify image data properties, model to define driving factors and main influencers and also anomaly detection cases.
Collection of different training runs for tracking and comparison can be collected to experiments. It collects hyperparameters and underlying metrics and you may compare the different runs.
MLFlow can also be used to compare and query among runs within an experiment. When you choose the best performing run you may save that as a model in your Fabric workspace.