MLCOST04-BP05 Use automated machine learning - Machine Learning Lens
Implementation guidanceResources

MLCOST04-BP05 Use automated machine learning

Automate your model development process by using systems that experiment with and select the best algorithms from high-performing options. These automated systems test various solutions and parameter settings to achieve optimal models, significantly speeding up development while reducing the need for manual experimentation and comparisons.

Desired outcome: You gain the ability to develop high-quality machine learning models in a fraction of the time traditionally required. By using automated machine learning tools like Amazon SageMaker AI Autopilot, you can focus on business problems rather than algorithm selection and parameter tuning. Your team can produce optimized models with better performance, reduce development costs, and accelerate time-to-market for ML-powered solutions.

Common anti-patterns:

  • Manually testing multiple algorithms and configurations one by one.

  • Spending excessive time on hyperparameter tuning without systematic approach.

  • Using the same algorithm for each problem without considering alternatives.

  • Neglecting cross-validation during model selection.

Benefits of establishing this best practice:

  • Dramatically reduced time to develop production-ready models.

  • Access to a broader range of algorithms and optimization techniques.

  • Improved model performance through systematic evaluation.

  • Lower costs through optimized resource utilization.

  • Ability for domain experts to build models without deep ML expertise.

Level of risk exposed if this best practice is not established: Medium

Implementation guidance

Automated machine learning (AutoML) systems democratize the process of building machine learning models. By automating key steps in model development—from data preparation to algorithm selection and hyperparameter tuning—these systems enable even those without extensive machine learning expertise to develop high-quality models.

When using AutoML solutions like Amazon SageMaker AI Autopilot, you provide your dataset and define your objective, and the system handles the complex work of exploring potential algorithms, optimizing parameters, and evaluating model performance. The system applies cross-validation procedures automatically to check that models can generalize well to new data. By ranking optimized models by their performance, AutoML can identify the best solution for your specific problem.

Beyond simply producing models, modern AutoML systems provide visibility into the development process, allowing you to understand what choices were made and why. This transparency builds trust in the models and provides learning opportunities for your team to understand what approaches work best for different problem types.

Implementation steps

  1. Evaluate your use case compatibility. Determine if your ML problem is suitable for automated machine learning solutions. AutoML works particularly well for standard machine learning tasks like classification, regression, and some time series forecasting scenarios.

  2. Prepare your data for AutoML. Clean your dataset, handle missing values, and convert categorical features appropriately. While AutoML handles feature engineering, providing high-quality data improves results. Use Amazon SageMaker AI Data Wrangler to simplify this preparation process.

  3. Set up Amazon SageMaker AI Autopilot with Canvas. Open Amazon SageMaker AI Canvas, import your dataset into Amazon S3, and configure to access this data. Define your target variable and specify your problem type (classification or regression) if known.

  4. Launch the automated ML job. Start Canvas training and let it analyze your data, select algorithms, and optimize models. Specify resources like maximum runtime and instance types to control costs. Canvas will automatically handle data preprocessing, feature engineering, algorithm selection, and hyperparameter optimization.

  5. Review candidate models. Examine the generated models along with their performance metrics. Autopilot provides detailed reports on the data exploration, feature engineering decisions, and model optimization steps it performed.

  6. Deploy the best model. Select the best-performing model from the Canvas recommendations and deploy it using Amazon SageMaker AI's deployment capabilities. You can deploy as a real-time endpoint or for batch inference depending on your needs.

  7. Monitor and evaluate performance. Set up model monitoring to track your model's performance in production and detect concept drift. Use Amazon SageMaker AI Model Monitor to automate this process.

  8. Customize and refine models. If needed, extract and customize the models generated by Autopilot. The solution provides full visibility into the notebooks and artifacts it creates, allowing you to further refine specific aspects of the model.

  9. Enhance model development with foundation models. Use Amazon Bedrock to incorporate foundation model capabilities into your AutoML workflow for tasks like text processing, content generation, and multimodal applications. Foundation models can complement traditional ML approaches handled by Autopilot.

  10. Use enhanced Canvas capabilities with Q integration. Use SageMaker AI Canvas with improved natural language support and Q integration for conversational data analysis, enabling business users to build models through natural language interactions.

  11. Implement intelligent preprocessing with generative AI. Use generative AI tools to enhance data preprocessing, augment training datasets, generate synthetic data for edge cases, and improve feature engineering through intelligent text and image processing.

Resources

Related documents:

Related examples: