How Booking.com modernized its ML experimentation framework with Amazon SageMaker

This article discusses how Booking.com modernized its machine learning (ML) experimentation framework by leveraging Amazon SageMaker to accelerate time-to-market for improved ML models.

Specifically, the article covers:

• Booking.com's challenges with long wait times for resources, lack of ML capabilities like hyperparameter tuning, and lengthy development cycles for ML models
• The approach to modernization by modifying existing code and developing a client package to interface with SageMaker APIs
• Configuring SageMaker pipelines with a config.ini file for customization
• SageMaker pipeline steps: data preparation, training, prediction, evaluation, condition checking, model packaging, and explainability
• Automatic model tuning with SageMaker's hyperparameter optimization strategies
• Model explainability using SageMaker Clarify for global and local interpretability
• Training optimization using SageMaker Debugger and TensorFlow Profiler
• Business outcomes: increased model training frequency, reduced failures, optimized training time, and advanced ML capabilities like experiment tracking