This is a set of instructions and exercises for a Machine Learning Workshop. This Machine Learning oriented content is focused on the use of Kubernetes (i.e. EKS).
These instructions assume your Workshop is using Event Engine. You will have the following resources pre-configured in us-west-2 (Oregon)
region:
- EKS Cluster named kf-sm-workshop
- Sagemaker Notebook with
AWS CLI
,eksctl
,kubectl
,aws-iam-authentictor
,git
, andkfctl
.
Note: if you are running this workshop on your own, please see the Self Paced Instructions (Note: as of April 30, these instructions are still in construction and may not work properly).
These are some things to be aware of before starting this workshop.
- Juypter Notebooks - This will be used in the KubeFlow labs and is an optional interface.
- JuypterLab - This is the recommended way to run the SageMaker labs. The multi-window capability makes the labs easier to run.
- Kubeflow Overview - Part of these instructions are to install and use Kubeflow on AWS EKS. Kubeflow on AWS EKS
- AWS EKS Architecture - The Managed Kubernetes Service from AWS.
- How to deploy Kubeflow on AWS EKS
- How to leverage the AWS Machine Learning Managed Service, Amazon SageMaker, from Kubeflow Pipelines.
- How to Build, Train, and Deploy Machine Learning Models using Amazon SageMaker.
- Batch Transform using Amazon SageMaker.
- Scaling Machine Learning Inference using SageMaker Multi-Model Endpoints.
- Best Practices for sizing Machine Learning instances.
- Machine Learning Model Monitoring (drift, re-training).
- Instance sizing
- A modern browser with an internet connection
- (recommended) 2 monitors or high enough resolution to run side-by-side windows
- (If running as part of an AWS Workshop) Amazon Chime App
- Login to your AWS Account using the supplied method.
- Navigate to SageMaker Service
- Verify / Change to the Oregon (us-west-2) region
- Launch Juypter (or Juypter Hub) on the BasicNotebookInstance
- Open a terminal and switch to 'bash' by typing
bash
at the terminal prompt - Run the command:
eksctl get clusters
- you should see the following:NAME REGION kf-sm-workshop us-west-2
- Run the command:
aws eks update-kubeconfig --name kf-sm-workshop
- Confirm connectivity to EKS by running
kubectl get nodes -A
- you should see a list of six nodes.
If you want to follow along in a different browser, navigate to The Source Github project.
There are several labs included with this Workshop, including: