This blog focuses on scaling many model training. While much of the buzz is around large model training, in recent years, more and more companies have found themselves needing to train and deploy many smaller machine learning models, often hundreds or thousands. Our team has worked with hundreds of companies looking to scale machine learning in production, and this blog post aims to cover the motivation and some best practices for training many models. Using the approaches described here, companies have seen order-of-magnitude performance and scalability wins (e.g., 12x for Instacart, 9x for Anastasia) relative to frameworks like Celery, AWS Batch, AWS SageMaker, Vertex AI, Dask, and more. While cutting edge applications of machine learning are leading to an explosion in model size, the need for many models cuts across industries.

Lessons from Kaggle competitions, including why XG Boosting is the top method for structured problems, Neural Networks and deep learning dominate unstructured problems (visuals, text, sound), and 2 types of problems for which Kaggle is suitable. Here is a summary of Anthony Goldbloom presentation at the Data Science Chicago Meetup, Nov 2 2015. Nice to see Anthony coming from financial statistics/econometrics (he mentioned his first job was with the Reserve Bank of Australia).