Machine Learning in Production
Summary
In this 1-day course, machine learning engineers, data engineers, and data scientists learn the best practices for managing the complete machine learning lifecycle from experimentation and model management through various deployment modalities and production issues. Students begin with end-to-end reproducibility of machine learning models using MLflow including data management, experiment tracking, and model management before deploying models with batch, streaming, and real-time as well as addressing related monitoring, alerting, and CI/CD issues. Sample code accompanies all modules and theoretical concepts.
Description
First, this course explores managing the experimentation process using MLflow with a focus on end-to-end reproducibility including data, model, and experiment tracking. Second, students operationalize their models by integrating with various downstream deployment tools including saving models to the MLflow model registry, managing artifacts and environments, and automating the testing of their models. Third, students implement batch, streaming, and real-time deployment options. Finally, additional production issues including continuous integration, continuous deployment are covered as well as monitoring and alerting.
By the end of this course, you will have built an end-to-end pipeline to log, deploy, and monitor machine learning models. This course is taught entirely in Python.
Duration
8 hours
Objectives
Upon completion, students should be able to:
- Track data and machine learning experiments to organize the machine learning life cycle
- Create, organize, and package machine learning projects with a focus on reproducibility and using a model registry to collaborate with a team
- Develop a generalizable way of handling machine learning models created in and deployed to a variety of environments
- Deploy basic CI/CD infrastructure using webhooks
- Explore the various production issues encountered in deploying and monitoring machine learning models
- Introduce various strategies for deploying models using batch, streaming, and real-time
- Explore various statistically rigorous solutions to drift and implement basic retraining methods
Audience
- Data Scientist
- Machine Learning Engineer
- Data Engineer
Prerequisites
- Intermediate experience using Python/pandas
- Working knowledge of machine learning and data science (scikit-learn, TensorFlow, etc.)
- Familiarity with Apache Spark
- Basic familiarity with object storage, databases, and networking
Outline
| Day #1 AM | ||
|---|---|---|
| Time | Lesson | Description |
| 40 min | Introductions, Setup & MLflow Overview | Registration, Courseware & Q&As |
| 20 min | Data Management | Design patterns to manage data lineage for data reproducibility |
| 10 min | Break | |
| 30 min | Experiment Tracking | Tracking experiments to organize the machine learning life cycle |
| 30 min | Advanced Experiment Tracking | Advanced methods for tracking ML experiments |
| 20 min | Model Management | Creating, organizing, and packaging machine learning projects with pre-processing code |
| 20 min | Model Registry | Artifact management for production models |
| 20 min | Webhooks and Testing | Integrating MLflow webhooks into the Model Registry |
| Day #1 PM | ||
|---|---|---|
| Time | Lesson | Description |
| 30 min | Production Issues | Various production issues encountered in deploying and monitoring machine learning models |
| 30 min | Batch Deployment | Various strategies for deploying models using batch including pure Python, Spark, and on the JVM |
| 10 min | Break | |
| 20 min | Streaming Deployment | How to perform inference on a stream of incoming data |
| 40 min | Real Time Deployment | Real time deployment with a focus on RESTful services |
| 10 min | Break | |
| 30 min | CI/CD | Continuous Integration, Continuous Deployment of ML models |
| 20 min | Drift Monitoring | Explore solutions to concept and data drift |
| 10 min | Alerting | Alerting strategies using email and REST integration |
Upcoming Classes
| Date | Time | Location | Price | |
|---|---|---|---|---|
| Sep 20 | 9:00 AM – 5:00 PM CEST | Online – Virtual – EMEA | $ 1000.00 USD | Register → |
| Oct 6 – 7 | 9:00 AM – 1:00 PM PDT | Online – Virtual – Americas (half-day schedule) | $ 1000.00 USD | Register → |
| Oct 20 | 9:00 AM – 5:00 PM CEST | Online – Virtual – EMEA | $ 1000.00 USD | Register → |
| Nov 17 – 18 | 9:00 AM – 1:00 PM PST | Online – Virtual – Americas (half-day schedule) | $ 1000.00 USD | Register → |
| Nov 24 | 9:00 AM – 5:00 PM CET | Online – Virtual – EMEA | $ 1000.00 USD | Register → |
| Dec 17 | 9:00 AM – 5:00 PM CET | Online – Virtual – EMEA | $ 1000.00 USD | Register → |
| Dec 28 – 29 | 9:00 AM – 1:00 PM PST | Online – Virtual – Americas (half-day schedule) | $ 1000.00 USD | Register → |