Coding as queues
ML workflows are not programs — they are pipelines. Treating each stage as a transform in a queue changes how you design for failure.
Lessons from the Tokens: short notes on AI, data science, machine learning, and research software. These are small lessons, observations, and patterns I want to keep track of and share.
ML workflows are not programs — they are pipelines. Treating each stage as a transform in a queue changes how you design for failure.
A short note on how leakage sneaks into model selection and how to keep your validation loop honest.
When you cannot get more data, think harder about the structure of the problem.