Software engineering patterns for machine learning applications (sep4mla)

摘要: To grasp the landscape of software engineering patterns for machine learning (ML) applications, a systematic literature review of both academic and gray literature is conducted to collect good and bad software-engineering practices in the form of patterns and anti-patterns for ML applications. From the 32 scholarly documents and 48 gray documents identified, we extracted 12 ML architecture patterns, 13 ML design patterns, and 8 ML anti-patterns. From these 33 ML patterns, we describe three major ML architecture patterns (“Data Lake”,“Distinguish Business Logic from ML Models”, and “Microservice Architecture”) and one ML design pattern (“ML Versioning”) in the standard pattern format so that practitioners can (re) use them in their contexts.