Architecting For Machine Learning on Amazon SageMaker
Welcome to the art and science of machine learning! During this 3-day course you will learn about the theory and application of machine learning in industry. This course is designed for architects and developers who did not previously have a background in AI/ML. You will spend 3 days performing data science tasks: training models, evaluating them, analyzing data, etc. After this 3 day period you will be better suited to design architecture guidelines for ML production applications. This course is also open to full-time data scientists, who will learn how to perform those tasks on AWS.
We will cover:
- Statistical machine learning
- Deep Learning
- Feature engineering
- Deploying a model into production
- Model evaluation and comparison
As a prerequisite to attending this course, we recommend reviewing Python programming using the statistical package Pandas. We also recommend having a Cloud Practiioner AWS Certification, but it is not required. Lastly, we recommend the book listed below. It is an excellent read, and clearly demonstrates all important concepts.
- https://pythonprogramming.net/data-analysis-python-pandas-tutorial-introduction/
- https://aws.amazon.com/certification/certified-cloud-practitioner/
- Deep Learning with Python by Francois Chollet
Agenda
Day One:
- Learn about ML on AWS
- Go through a sample lab
- Break into teams and focus on a new machine learning project
Deliverable: Produce a sample writeup explaining your modeling strategy
Day Two:
- Learn about feature engineering on AWS
- Start new notebooks, sample your code, and develop preliminary data sets
- Read the evaluation questions, and begin to think about how your modeling strategy compares to the evaluation questions.
- Finish most of your feature engineering.
Deliverable: Product the first version of your trained model
Day Three:
- Learn about putting your model into production.
- Evalute your project against other approaches
- Design a reference architecture demonstrating your final solution Stretch goal: Produce multiple versions of your model and compare them
What you'll need
- AWS Account log in credentials
- Github account to share code with your project partners
- Kaggle account to download data sets
We're assuming that you will complete this course using an AWS account we will provide you with throughout the course.