Presented at AWS Community Summit: Getting started with Amazon SageMaker.
Related Block post - https://medium.com/peak-product/deploy-sagemaker-models-on-aws-adf41e1a1853
Michael Pearce, DevOps Engineer @ Peak AI.
Disclaimer: Due to the fast moving nature of AWS, may be out of date!
2. ■ Michael Pearce
■ IT/DevOps Team Leader @ Peak
■ Based in Manchester
■ 5x AWS certified, 2x Linux
About Me
–
peak.ai
3. ■ The UK’s leading enterprise AI company, founded in
November 2014
■ 70 employees with plans to double in size this year
■ Offices in Manchester, Jaipur, London, Brisbane and
Edinburgh
■ Venture-backed, with a total of £11m raised to date
from leading UK investors. Consistent revenue
growth of 250% year-on-year
■ ML Competency Partner
peak.ai
About Peak
4. Do great things with data
30%
Revenue growth Profit margins
50%
AI-powered businesses But it’s not easy
Complex technology Skills are scarce
peak.ai
Peak believes that every business must
become AI-driven in order to thrive and
succeed in the modern world…
We provide businesses with the
technology and the skills they need to
become AI-driven to compete.
5. Multiple Business Systems
Can be inflexible, slow and
expensive to maintain
Data Silos
Data is in silos, caused by
proliferation of cloud applications
Nowhere to Build AI
Data warehouses are not built to
train algorithms
Business systems are not built for AI
peak.ai
6. Think of it like an adaptive brain - powering
every aspect of an enterprise, simultaneously.
Ingest data from
any source
Train machine
learning models
Transform and
unify data
Connect with
other systems
So... Peak built the first Enterprise AI System
7. ■ A managed service to Build, Train, and Deploy
machine learning on AWS
■ Off the shelf algorithm, or build your own
■ Really does simplify and speed up the infrastructure
needed for machine learning
○ Prepackaged Images
○ Out Of Box algorithms or your own
■ Developing rapidly!
What is SageMaker?
–
peak.ai
11. ■ Build your model (technically optional)
■ Train it (also optional)
■ Save it to s3 (optional)
■ Write the code to build your API
○ Create a template
○ Containerise it
peak.ai
Development
13. ■ Model
■ Endpoint Configuration
■ Endpoint
■ First Step - Create the Model
○ Provide the location of the model artifacts
and inference code.
Setting up the Endpoint
–
peak.ai
15. ■ Create the Model
■ Create the Model Configuration
○ Select the model
○ Instance Type
○ How many Instances?
○ Elastic Inference?
■ Apply to model configuration to a Endpoint
SageMaker: Endpoint
–
peak.ai
21. ■ Input values - Mostly static and/or implied
○ Network configuration
○ Instance counts, size etc.
○ Conditional autoscaling targets (based on
instance type)
■ Use the SDK
○ Automate the build process in the
background
○ Build into a user friendly interface
peak.ai
Making It Scale
25. PROS
■ Provides Scalability, High throughput, High
reliability
■ Enables A/B Testing
CONS
■ Long convoluted endpoint to call with aws service
proxy or sdk
■ Lots of moving parts
■ Expensive!
peak.ai
Hosting Endpoint
27. SageMaker is…
■ A managed service to Build, Train, and Deploy
machine learning on AWS
■ Moving fast!
We looked at…
■ Packaging your ML model
■ Setting up SageMaker Endpoints
○ Autonomously
○ With a friendly UI
■ Hosting the models on the Endpoints
○ Making it Production ready
○ Abstracting away the complicated moving
parts
Overview
–
peak.ai