Amazon SageMaker
For MLOps
Aman Srivastava
Lack of etiquette and manners is a huge turn off.
KnolX Etiquettes
 Punctuality
Join the session 5 minutes prior to the session start time. We start on
time and conclude on time!
 Feedback
Make sure to submit a constructive feedback for all sessions as it is very
helpful for the presenter.
 Silent Mode
Keep your mobile devices in silent mode, feel free to move out of session
in case you need to attend an urgent call.
 Avoid Disturbance
Avoid unwanted chit chat during the session.
1. What is SageMaker
 Introduction to Amazon SageMaker
2. SageMaker for MLOps
 Introduction to MLOps
 SageMaker Workflows
 AirFlow & StepFunction
3. CodePipeline in SageMaker
§ Accelerate Model Development
§ Deploy & Manage Models in Production
4. Customize Infrastructure
 Automate custom integration and deployment
5. Quality Control
 Data Monitoring
 Model Monitoring
6. Containerization in SageMaker
 Pre-built SageMaker Images
 Building your own container
7. Demo
8. Q&A
Introduction to Amazon SageMaker
 SageMaker is an integrated platform that combines Amazon SageMaker's machine learning capabilities with
DevOps practices.
 Enabling streamlined development, deployment, and management of machine learning models at scale.
Train & Evaluate Model
Create Endpoint
Configuration
Monitor &
Maintain Endpoint
Deploy the Model
Create SageMaker Model Update or Delete Endpoint
SageMaker for MLOps
02
Introduction to MLOps
• Problem Definition and Planning
• Data Collection and Preparation
• Model Development
• Version Control
 MLOps = ML + DEV + OPS
 MLOps is the seamless integration of Machine Learning (ML), Software Development (DEV), and Operations
(OPS) to streamline the end-to-end lifecycle of deploying and managing ML models.
Development and
Planning
CI/CD
Monitoring and
Management
Governance and
Collaboration
• Continuous Integration (CI)
• Model Deployment
• Continuous Deployment (CD)
• Monitoring and Logging
• Feedback Loop and Model Updating
• Scaling and Resource Management
• Security and Compliance
• Documentation
• Collaboration and
Communication
SageMaker Workflows
 A tool designed for constructing and overseeing machine learning pipelines.
 Utilize SageMaker custom operators on your Kubernetes cluster and components for Kubeflow Pipelines.
 Execute on-demand or scheduled non-interactive batch runs of your Jupyter notebook.
 Leverage SageMaker APIs to export configurations for the creation and management of Airflow workflows.
 Develop multi-step ML workflows in Python orchestrating SageMaker infrastructure without the need for
separate resource provisioning.
AirFlow & StepFunction
 Airflow: Orchestrate and automate complex workflows in the cloud with Apache Airflow on Amazon Web
Services.
 StepFunction: Serverless orchestration for integrating Lambda functions and AWS services in event-driven
workflows with a graphical console view.
Batch Inference at Scale with Amazon SageMaker Task Execution using Lambda
CodePipeline in SageMaker
03
Accelerate Model Development
 Standardizing ML development environments with Amazon SageMaker Projects accelerates innovation,
simplifying project launches through consistent setups and templates for tested tools and libraries.
 This boosts data scientist productivity by providing standardized environments with code templates, source
control, and CI/CD pipelines.
Deploy & Manage Model in Production
 Quickly reproduce your models for troubleshooting.
− Amazon SageMaker logs each workflow step, creating an audit trail of model artifacts for easy
reproduction and debugging of potential issues in production models.
 Centrally track and manage model versions.
− SageMaker Model Registry streamlines ML app development, centralizing model versions and metrics
for easy deployment selection and automated audit trail of approval workflows.
04
Automate custom integration and deployment
 SageMaker Projects accelerates ML model deployment with CI/CD, ensuring agility in production.
 SageMaker Projects integrates ML development, boosting agility through CI/CD, version control, A/B testing,
and automation for swift model deployment.
04
Data Monitoring
 Amazon SageMaker Model Monitor ensures the ongoing quality of machine learning models in production
through continuous monitoring via real-time endpoints or scheduled batch transform jobs.
 Model Monitor alerts on model quality deviations, enabling proactive actions like retraining without manual
monitoring. Use prebuilt features without coding or code for custom analysis and flexible model monitoring.
Model Monitoring
 Model quality monitoring jobs assess model performance by comparing its predictions to actual Ground Truth
labels, merging real-time or batch inference data with stored labels in an Amazon S3 bucket for comparison.
Image Attribute: Image sourced from AWS (Amazon Web Services) documentation on SageMaker. Source: Link
05
Pre-built SageMaker Images
 Amazon SageMaker provides containers for built-in algorithms and pre-built Docker images for popular ML
frameworks (e.g., MXNet, TensorFlow, PyTorch) and supports libraries like scikit-learn and SparkML.
Building your own container
 Create a custom container if you're building or training a personalized model with a framework that lacks a
pre-built image.
Image Attribute: Image sourced from AWS (Amazon Web Services) documentation on SageMaker. Source: Link
06
07
Amazon SageMaker for MLOps Presentation.

Amazon SageMaker for MLOps Presentation.

  • 1.
  • 2.
    Lack of etiquetteand manners is a huge turn off. KnolX Etiquettes  Punctuality Join the session 5 minutes prior to the session start time. We start on time and conclude on time!  Feedback Make sure to submit a constructive feedback for all sessions as it is very helpful for the presenter.  Silent Mode Keep your mobile devices in silent mode, feel free to move out of session in case you need to attend an urgent call.  Avoid Disturbance Avoid unwanted chit chat during the session.
  • 3.
    1. What isSageMaker  Introduction to Amazon SageMaker 2. SageMaker for MLOps  Introduction to MLOps  SageMaker Workflows  AirFlow & StepFunction 3. CodePipeline in SageMaker § Accelerate Model Development § Deploy & Manage Models in Production 4. Customize Infrastructure  Automate custom integration and deployment 5. Quality Control  Data Monitoring  Model Monitoring 6. Containerization in SageMaker  Pre-built SageMaker Images  Building your own container 7. Demo 8. Q&A
  • 5.
    Introduction to AmazonSageMaker  SageMaker is an integrated platform that combines Amazon SageMaker's machine learning capabilities with DevOps practices.  Enabling streamlined development, deployment, and management of machine learning models at scale. Train & Evaluate Model Create Endpoint Configuration Monitor & Maintain Endpoint Deploy the Model Create SageMaker Model Update or Delete Endpoint
  • 6.
  • 7.
    Introduction to MLOps •Problem Definition and Planning • Data Collection and Preparation • Model Development • Version Control  MLOps = ML + DEV + OPS  MLOps is the seamless integration of Machine Learning (ML), Software Development (DEV), and Operations (OPS) to streamline the end-to-end lifecycle of deploying and managing ML models. Development and Planning CI/CD Monitoring and Management Governance and Collaboration • Continuous Integration (CI) • Model Deployment • Continuous Deployment (CD) • Monitoring and Logging • Feedback Loop and Model Updating • Scaling and Resource Management • Security and Compliance • Documentation • Collaboration and Communication
  • 8.
    SageMaker Workflows  Atool designed for constructing and overseeing machine learning pipelines.  Utilize SageMaker custom operators on your Kubernetes cluster and components for Kubeflow Pipelines.  Execute on-demand or scheduled non-interactive batch runs of your Jupyter notebook.  Leverage SageMaker APIs to export configurations for the creation and management of Airflow workflows.  Develop multi-step ML workflows in Python orchestrating SageMaker infrastructure without the need for separate resource provisioning.
  • 9.
    AirFlow & StepFunction Airflow: Orchestrate and automate complex workflows in the cloud with Apache Airflow on Amazon Web Services.  StepFunction: Serverless orchestration for integrating Lambda functions and AWS services in event-driven workflows with a graphical console view. Batch Inference at Scale with Amazon SageMaker Task Execution using Lambda
  • 10.
  • 11.
    Accelerate Model Development Standardizing ML development environments with Amazon SageMaker Projects accelerates innovation, simplifying project launches through consistent setups and templates for tested tools and libraries.  This boosts data scientist productivity by providing standardized environments with code templates, source control, and CI/CD pipelines.
  • 12.
    Deploy & ManageModel in Production  Quickly reproduce your models for troubleshooting. − Amazon SageMaker logs each workflow step, creating an audit trail of model artifacts for easy reproduction and debugging of potential issues in production models.  Centrally track and manage model versions. − SageMaker Model Registry streamlines ML app development, centralizing model versions and metrics for easy deployment selection and automated audit trail of approval workflows.
  • 13.
  • 14.
    Automate custom integrationand deployment  SageMaker Projects accelerates ML model deployment with CI/CD, ensuring agility in production.  SageMaker Projects integrates ML development, boosting agility through CI/CD, version control, A/B testing, and automation for swift model deployment.
  • 15.
  • 16.
    Data Monitoring  AmazonSageMaker Model Monitor ensures the ongoing quality of machine learning models in production through continuous monitoring via real-time endpoints or scheduled batch transform jobs.  Model Monitor alerts on model quality deviations, enabling proactive actions like retraining without manual monitoring. Use prebuilt features without coding or code for custom analysis and flexible model monitoring.
  • 17.
    Model Monitoring  Modelquality monitoring jobs assess model performance by comparing its predictions to actual Ground Truth labels, merging real-time or batch inference data with stored labels in an Amazon S3 bucket for comparison. Image Attribute: Image sourced from AWS (Amazon Web Services) documentation on SageMaker. Source: Link
  • 18.
  • 19.
    Pre-built SageMaker Images Amazon SageMaker provides containers for built-in algorithms and pre-built Docker images for popular ML frameworks (e.g., MXNet, TensorFlow, PyTorch) and supports libraries like scikit-learn and SparkML.
  • 20.
    Building your owncontainer  Create a custom container if you're building or training a personalized model with a framework that lacks a pre-built image. Image Attribute: Image sourced from AWS (Amazon Web Services) documentation on SageMaker. Source: Link
  • 21.
  • 22.