MLOps with TFX — DevFest Nairobi 2022

A lot of developers see TensorFlow from the experimental perspective: tensorflow machine learning algorithms, neural networks in tensorflow, tensorflow hub for datasets and reusable models, tensorboard for visualization and experimentation, etc. What they are probably yet to know is how much TensorFlow now offers regarding model deployment and management in production. In reality, TensorFlow was also built with MLOps in mind.

In the early days, while TensorFlow offered flexibility, it sort of lacked a complete end-to-end production system.

Sibyl (now TFX), on the other hand, had robust end-to-end capabilities, but lacked flexibility.

This talk is a dive into TFX and a little beyond deploying TensorFlow models for production.

Transcript

1. Nairobi MLOps with TFX (TensorFlow Extended) Olayinka Peter Oluwafemi GDE, Machine Learning
2. MLOps with TFX (TensorFlow Extended)
3. The Discipline of ML Engineering Programming took form over the decades and matured into the discipline of Software Engineering. Can we say the same for applied ML yet? No? Why?
4. “But Software Engineering Principles Already Can Suffice?” Well, someone could say this. But, extended forms of software engineering have their own “engineering” procedures. E.g, Mobile Engineering.
5. The “Smooth Sail” of Applied ML in Prod? While ML Algorithms are important, they are usually insufficient in achieving the successful application of ML in a single product. There are more essential aspects in the ML life cycle.
6. “Sibyl”, the birth of TFX In 2007, a team of Googlers built Sibyl, a platform with the purpose of enabling massive-scale ML, catered to production use. What Sibyl focused on were tools for several aspects of the ML workflow including Data Ingestion, Data Analysis and Validation, Training, Model Analysis, and Training-Serving Skew Detection.
7. TensorFlow 🎉 Fast-forward to 2015 when Google announced the public release of TensorFlow. The superpower of TensorFlow is its flexibility, which allowed it to be used for a lot more than DL and its popularity in both research and production positioned it as the go-to framework for authoring ML algorithms.
8. Early TensorFlow Models —> Production :\ In the early days, while TensorFlow offered flexibility, it lacked a complete end-to-end production system. Sibyl (now TFX), on the other hand, had robust end-to-end capabilities, but lacked flexibility.
9. Come in, TFX That tiny problem of early TensorFlow became very apparent in no time — the need for an end-to-end ML platform for TensorFlow to accelerate ML - even within Google. And then, in 2017, TFX was launched within Google.
10. ML Engineering; The Idea “On any given day there are thousands of TFX pipelines running, which are processing exabytes of data and producing tens of thousands of models, which in turn are performing hundreds of millions of inferences per second.” - the TFX team within Google, 2019
11. ML Engineering; The Idea Upon that internal success, and the supposition that the same idea of ML engineering will be needed by organizations and individuals everywhere in the world, the team decided to publicise their idea of ML Engineering and TFX.
12. ML Engineering; The Principles 1. Start with simple rules and heuristics 2. Move to simple ML (i.e., simple models) and realize large gains 3. Move to ML with more features and more advanced models to realize decent gains. 4. Move to state-of-the-art ML, manage refinement and complexity (for solutions to the problems that are worth it), and realize small gains. 5. Apply the above launch-and-iterate cycle to more aspects of products and to solve more problems.
13. What TFX is TFX is an end-to-end platform for deploying production ML pipelines. It provides a configuration framework and shared libraries to integrate common components needed to define, launch, and monitor your machine learning system.
14. Getting Started with TFX tensorflow.org/tfx
15. En conclusión Just as code and data evolve over time, models also evolve over time.
16. Thank you @olayinkapeter_