Building Computational Graphs Across Multiple Open Source Frameworks
Modern machine learning (ML) teams have found great success with combining multiple ML models as ensembles for better predictive performance. However, algorithms implemented in different ML frameworks can’t be serialized into a single combined model for deployment. In this talk, we share how DoorDash uses a computational graph approach via a domain specific language (DSL) to allow teams to use multiple frameworks at once in a single combined model. We demonstrate how we generate and interpret instances of this DSL to allow our teams to quickly define and deploy customized models and get high runtime performance during inference.
3 key takeaways:
Interoperability: Support for multiple frameworks lets teams develop models in areas of their expertise, rather than having to conform to a standard.
Accuracy: Unlocking a broad range of ensemble methods should enable greater predictive performance.
Performance: The DSL approach makes it possible to unify model implementations in C++ which will give better performance.
Hebo is a Machine Learning Platform Engineer at DoorDash, primarily working on model training and management tooling and pipelines. Previously he has worked on data ETL, high-available distributed systems, and machine learning-based cyber security detection cloud service. Hebo graduated from Columbia University with an MS in Computer Science and Applied Mathematics.