MLOps คืออะไร ? ทําไมคนถึงวาดี Kamolphan Liwprasert (Fon)

about:me FON Kamolphan Liwprasert Senior Machine Learning Engineer @Sertis Master in Computer Science student @GeorgiaTech Road to Data Engineer instructor @DataTH

MLOps คืออะไร ?

ML Machine Learning Operations Ops https://neptune.ai/blog/end-to-end-mlops-platforms

https://youtu.be/VfcY0edoSLU นิยาม AI System แบบงาย ๆ AI System = * Software Engineering DevOps Data Engineering Data Science ML/AI Research ML Engineering Code + Model + DATA

MLOps แตกตางจาก DevOps อยางไร? DevOps MLOps

3 Levels of ML Software 1. Data ▸ Data Engineering Pipelines ▸ Data Quality is a key 🔑 2. Model ▸ Machine Learning Pipelines ▸ Train, Evaluate, Test 󰙤 3. Code ▸ Model Serving & Predictions ▸ Deployment strategies & Infra 🐳☁

MLOps Continuous Training Continuous Monitoring CI/CD Concept drift Logging Canary Deployment A/B Testing Data-Centric AI vs Model-Centric AI Monitoring Healthcheck Testing Data Version Control Experiment Tracking Data Lineage Lifecycle Automation Data Quality Data Labeling Data Augmentation Train & Evaluate Model Versioning Model Serving & Deployment Optimization Error Reporting

ทําไม ตองมี MLOps ? ? ? ?

ทําไมคนเพิ่งมาพูดถึง MLOps

https://cloud.google.com/solutions/machine-learning/mlops-continuous-delivery-and-automation-pipelines-in-machine-learning Hidden Technical Debt in Machine Learning Systems Elements for ML systems

Rule of Machine Learning Martin Zinkevich https://developers.google.com/machine-learning/guides/rules-of-ml

To make great products: do machine learning like the great engineer you are, not like the great machine learning expert you aren’t. Martin Zinkevich https://developers.google.com/machine-learning/guides/rules-of-ml

1 2 3 4 Benefit of MLOps Automate process ชวยลด Technical Debt ในระยะยาว ได Pipeline ที่ดี ทําให data scientist ทํางานรวมกันไดงาย ชวยลดความเสี่ยง และขอผิด พลาด human error ปรับปรุง และ Maintain ได งาย สามารถ scale ได

เริ่มตนกับ MLOps ยังไงดี?

Tools DVC (Data Version Control)

Cloud Platforms Azure Machine Learning Amazon SageMaker Google Cloud AI Platform

Concerns ● แตละ tool มี usecase ที่เหมาะสมของเครื่องมือแตละอยาง ⚒ ● Pricing $$$$ บาง tool มีคาใชจายคอนขางสูง 💸 ● Over-engineering : Perfect is the enemy of good

เริ่มตนกับ MLOps - Beginner edition Empathize เรียนรูจาก pain Git VersionControl Docker Kubernetes CI/CD Programming!

RECAP

https://ml-ops.org/content/mlops-principles

References & Resources ● MLOps https://ml-ops.org ● Introducing MLOps, Mark Treveil and team (O’Reilly Media). ● Rules of Machine Learning, Martin Zinkevich https://developers.google.com/machine-learning/guides/rules-of-ml ● Hidden Technical Debt in Machine Learning Systems https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf ● What Is MLOps?, Nvidia https://blogs.nvidia.com/blog/2020/09/03/what-is-mlops/ ● A Chat with Andrew on MLOps: From Model-centric to Data-centric AI, Andrew Ng https://youtu.be/06-AZXmwHjo ● Let’s talk about MLOps, Christian Barra https://youtu.be/K5x6dxjY1vA ● MLOps: Continuous delivery and automation pipelines in machine learning, Google Cloud https://cloud.google.com/solutions/machine-learning/mlops-continuous-delivery-and-automation -pipelines-in-machine-learning ● Awesome-mlops, visenger (on GitHub) https://github.com/visenger/awesome-mlops ● CML, powered by DVC https://cml.dev https://dvc.org

Q & A 💬

Thank you 😃 Feedback is a gift 🎁 bit.ly/feedback-form-fon Sertis is hiring 󰠁 bit.ly/sertis-apply-now Road to Data Engineer course 🎓 bit.ly/road2de-datatalk *ลด 5% ถึงเที่ยงคืนวันที่ 28 มี.ค.นี้