Machine Learning | Membangun Sistem Machine Learning

Sesi ini akan membahas konsep dasar MLOps (Machine Learning Operations) dan mengapa peranannya sangat penting dalam siklus hidup machine learning modern. Peserta akan dikenalkan pada prinsip-prinsip utama MLOps seperti reproducibility, automation, monitoring, dan versioning. Selanjutnya, akan dijelaskan kerangka MLOps Maturity Level untuk memahami sejauh mana kematangan sistem ML yang dimiliki, mulai dari proses manual hingga pipeline otomatis. Pada bagian akhir, peserta akan diajak mengeksplorasi bagaimana cara meningkatkan level kematangan MLOps menggunakan tools open-source seperti MLflow untuk experiment tracking, GitHub sebagai kontrol versi dan kolaborasi tim, serta Docker untuk memastikan reproducibility dan portability. Dengan mengikuti sesi ini, peserta diharapkan mampu membangun landasan kuat dalam pengembangan sistem ML yang andal, scalable, dan siap produksi.

Transcript

1. Maybe you need to stop build model and learn to maintain your system. Hot Course Pelajari cara menerapkan model machine learning pada tahap produksi dengan membuat dan mengimplementasikan prinsip MLOps menggunakan mlflow.
2. Membangun Sistem Machine Learning Machine Learning Mochamad Rafy Ardhanie Curriculum Developer
3. Key Point Machine Learning • What is MLOps and Why It Matters • Introduce key MLOps principles • MLOps maturity • Climbing MLOps maturity levels with open-source tools
4. Quick Questions
5. Flashback~ Youʼve trained a great machine learning model. Youʼve validated it, tuned it, and your accuracy is excellent. But hereʼs the real question... Now what? Machine Learning Life Cycle
6. Many machine learning projects never make it past the notebook. Not because the models are bad, but because we donʼt have the right infrastructure, process, or mindset to take them further. Thatʼs where MLOps comes in. Gartner reports that 50% of AI projects never make it to production due to poor deployment infrastructure¹. McKinsey found that AI models in production degrade by 1530% yearly without proper MLOps². RAND Corporation notes that companies often lack data pipelines and monitoring tools to keep AI running effectively³. The Silent AI Killer: Lack of MLOps & Infrastructure For Your Information
7. ML Life Cycle Is All You Need
8. ML Life Cycle Isnʼt All You Need
9. What is MLOps and Why It Matters Machine Learning
10. Fun Fact from DevOps
11. Start with DevOps DevOps is a set of practices that combines software development Dev and IT operations Ops to shorten the development lifecycle and deliver high-quality software continuously. It emphasizes collaboration, automation, and monitoring across all stages of software delivery — from coding to deployment.
12. DevOps vs MLOps
13. Most of CS Students
14. ML System Workflow
15. Please Welcome, MLOps! MLOps Machine learning Operations) is a set of practices that helps data scientists and ML engineers to manage the machine learning ML life cycle more efficiently. It aims to bridge the gap between development and operations for machine learning.
16. Why Do We Needs MLOps? We need to retrain our machine learning model to ensure it remains accurate and relevant as new data patterns, user behavior, or real-world conditions change over time.
17. Why ML Engineer team need MLOps? 4 Common Problems with Scaling ML Getting Started Mundane Steps with lots of boilerplate Iteration Speed Slow, redundant iterative development, long time-to-market. Collaboration Team Growth, Different skills and backgrounds Productionisation Handoff to production is manual and bottlenecked, difficulty to reproduce and control experiments
18. An Example End-to-End MLOps Platform Implementation
19. Quiz Time What is the main goal of MLOps in a machine learning workflow? A. To improve model accuracy by using deeper neural networks B. To automate and manage the lifecycle of machine learning models C. To collect more data from users automatically D. To replace data scientists with AI tools
20. Introduce Key MLOps principles Machine Learning
21. Reproducibility 01 Guarantees that experiments can be reliably repeated using the same code, data, and configurations. Versioning 02 Tracks and manages different versions of datasets, models, and pipelines for better traceability and rollback. Testing 03 Validates the quality and reliability of ML models and pipelines through systematic testing. Automation 04 Streamlines ML workflows such as training, evaluation, and deployment to reduce manual effort and errors. Monitoring 05 Continuously observes model performance and system behavior to detect issues like data drift or degradation. Security 06 Protects data, models, and infrastructure from unauthorized access and malicious threats. Beyond the MLOps principles above, there may be some differences in perspective depending on the reading sources you find. MLOps Principles
22. Building Machine Learning System We need the experimentation stage to build the right model, and the production stage to run the model right. Production Stage Experimentation Stage
23. Quiz Time Why is it important to separate the experimentation stage from the production stage in MLOps? A. To prevent developers from accessing model code B. To reduce the cost of training models in production C. To allow safe testing and iteration without affecting live systems D. To make model training slower and more controlled
24. MLOps maturity Machine Learning
25. MLOps Maturity Level Google Almost every company wants to be a Data + AI company, but the majority are new to Machine Learning.
26. MLOps Maturity Level Microsoft
27. Continuous Training This is the level of maturity where we implement automated Continuous Training. MLOps Maturity Google Level 1
28. Climbing MLOps maturity levels with open-source tools Machine Learning
29. MLflow MLflow is an open-source platform that helps manage the complete machine learning lifecycle, including experimentation, reproducibility, deployment, and model registry.
30. GitHub Actions GitHub Actions is a CI/CD tool that lets you automate workflows like building, testing, and deploying code directly from your GitHub repository.
31. Docker Docker is a platform that enables you to package applications and their dependencies into lightweight, portable containers that can run consistently across different environments.
32. Three Tools is All You Need* Assist machine learning practitioners and teams in handling the complexities of the machine learning process. MLFlow GitHub is a proprietary developer platform that allows developers to create, store, manage, and share their code. GitHub Docker Docker is a software platform that allows you to build, test, and deploy applications quickly. Connections is a key! *This is for open-source tools. You can scale up with using another services tools like Kubeflow, Azure Machine Learning, AWS Sagemaker, etc.
33. Finalize with Prometheus and Grafana
34. Final Look • MLflow tracks the "what" and "how well" of your models. • GitHub Actions automates "when" and "how" things should happen. • Docker ensures "where" and "with what" your code and models run.
35. Quiz Time What is the primary role of MLflow in an MLOps workflow? A. Serving as a cloud platform to deploy containers B. Managing version control for source code C. Tracking experiments and managing the ML model lifecycle D. Automating hyperparameter tuning using reinforcement learning
36. Code for Lyfe
37. Maybe you need to stop build model and learn to maintain your system. Hot Course Pelajari cara menerapkan model machine learning pada tahap produksi dengan membuat dan mengimplementasikan prinsip MLOps menggunakan mlflow.
38. Thank You Rafyardhani Rafy@dicoding.com Rafyardhani RafyArdhanie Machine Learning
39. Feedback! dicoding.id/devcoachfeedback