The toolkit can be used to help data science teams be productive by automating various steps of the data science workflow – including feature engineering, hyperparameter tuning, model search, and deployment – for a fully controlled and transparent augmented ML experience. This is available in Databricks Labs custom solutions for citizen and expert data scientists. AutoML Toolkit executions are automatically tracked in MLflow.
The Databricks Labs project is an experimental end-to-end supervised learning solution for automating the steps like feature clean-up, feature vectorization, model selection and training, hyper parameter optimization and selection, batch prediction and logging of model results and training runs.
The Unified Analytics Platform includes three main components:
- Databricks Workspace: With the goal of unifying data science and engineering, the workspace handles all analytic processes (from ETL to model training and deployment), leveraging shared interactive notebooks, tools, and APIs.
- Databricks Runtime: The runtime component helps with data preparation and continuously trains and deploys the models for AI/ML applications. It supports integrations between Hyperopt, MLlib, and MLflow, which enables distributed conditional hyperparameter tuning, automated tracking, and enhanced visualizations. The users can get started with pre-configured clusters including some of the popular ML frameworks like Hadoop, Kafka, Spark, Parquet, TensorFlow, Keras, and Scikit Learn.
- Databricks Cloud Service: The cloud service helps with managing the infrastructure complexity by offering a fully managed service on the cloud. The cloud offerings include Microsoft Azure integration and Amazon Web Services (AWS).