Ask HN: What are good patters / practices in Machine Learning Development?

[snowpid]

Hello,

Currently I am somehow experienced Machine Learning Engineer. I'm trying to improve my Software Engineering skills in Machine Learning by using more good practices / patterns. I think using pytorch lightning is a good starting point as the API forces you to use good practices. But I like to understand more. (I am aware ML is changing fast but maybe you see some longterm patterns)

1.) What do you think are good practices in ML? 2.) Can you recommend some books? I found " Designing Machine Learning Systems: An Iterative Process for Production-Ready Applications " and " Machine Learning Design Patterns". What do you think about them?

Thank you!

[alexmolas]

1 - the first rule of ML is to start without ML.

2 - do ML like the great engineer you are, not like the great ML expert you aren’t.

[chuckhend]

Most of the content in https://www.oreilly.com/library/view/designing-data-intensiv... applies to ML systems.

https://github.com/noahgift also has a lot of content that is worth following.

[snowpid → chuckhend]

Have you read the book? What is your experience?

[chuckhend → snowpid]

It was required reading at a place I worked a few years back. The chapters on batch and stream processing are good. Both applicable to batch/streaming inference.

[vietthan]

This was posted a few days ago actually

https://eugeneyan.com/writing/llm-patterns/