Home

Awesome

ML Operationalization Starter Kit for ML Operationalization

Contents

  1. ML Operationalization
  2. Key Capabilities needed for Operationalizing Models
  3. Frameworks for Operationalizing Models
  4. IBM Cloud Pak For Data for ML Operationalization
  5. ML Ops Starter Kit
  6. ML Ops Starter Kit Projects
  7. Where to use ML Ops Starter Kit Projects
  8. ML Ops Starter Kit Roadmap
  9. Contents of the Repository
  10. How to get Started with this Starter Kit
  11. Contributors

ML Operationalization<a class="anchor" id="ml-ops-intro">

ML Operationalization refers to operationalization of Machine Learning Models for production use to realize business value out of those Models.

ML Operationalization covers standard ML Lifecycle (CRISP-DM) overlaid with Continuous Integration/Continuous Development paradigm. It covers Data Priovisioning & Governance, Data Preparation for Model Development, Model Development, Model Validation & Governance, Infusing Model's Insight in Application and Model Monitoring in a continuous fashion.

In contrast to Appplication DevOps, MLOps not only takes care of Continuous Integration and Continuous Deployment but also covers Continuous Training, Continuous Validation and Continuous Monitoring.

The diagram below show the ML Ops steps and persona

For more infornation on conceptual view of ML Operationalization please check Operationalizing AI

Key Capabilities needed for Operationalizing Models<a class="anchor" id="ml-ops-capabilities">

Here are the key Functional Capabilities those are needed in establishing ML Operationalization in an organization.

Along with these functional capabilities having right set of Non Functional Capabilities is important too

Frameworks for Operationalizing Models<a class="anchor" id="ml-ops-cpd-intro">

There are various frameworks and platforms available for ML Operationalization with varied degree of support for required Functional and Non Functional capabilities

IBM Cloud Pak For Data for ML Operationalization <a class="anchor" id="ml-ops-using-cpd-">

IBM Cloud Pak For Data is an End to End platform that supports required Functional and Non Functional Capabilities required by ML Operationalization in a comprehensive manner. It is a platform to accelerate realization of business value from Machine Learning Models with an open, extensible data and AI platform that runs on any cloud.

For detailed information about Cloud Pak For Data please check here

ML Ops Starter Kit<a class="anchor" id="ml-ops-starter-kit">

ML Ops Starter Kit is an End 2 End framework to help one get started with ML Operationalization. It ruses components of Cloud Pak For Data. However, can be extended for any other Open Source of Propreitory ML Ops frameworks too for supporting any kind of hybrid approach.

ML Ops Starter Kit Projects<a class="anchor" id="ml-ops-sk-projects">

ML Ops Start Kit contains a set of Get Started Projects. Each Project has Instructions, Code/Notebooks, and Data packaged together in a single asset (.zip file) that can be imported as new 'Project' in Cloud Pak For Data project in a single click.

ML Ops Start Kit uses Customer Churn Use case for a communications company. The use case covers key aspects of operationalizing a customer churn model like

+ Merging usage data with demographic data

+ Automatically selecting algorithms

+ Adding mathematical features if needed

+ Monitoring the model for bias around age or gender

+ Checking the model's explanations on the key factors that might create customer churn, such as income or dropped calls

Each Get Started project supports a specific Machine Learning framework. Right now there are 3 Get Started projects supporting + Scikit Learn + Spark MLLib + AutoAI + More frameworks; R, Tensore Flow/Keras, SPSS Modeler, etc; are coming soon.

Where to use ML Ops Starter Kit Projects<a class="anchor" id="ml-ops-sk-projects-usage">

ML Ops Starter Kit works on any Cloud Pak For Data cluster (that has necessary components) including Cloud Pak For Data Experience.

ML Ops Starter Kit can be potentially extended to other ML Ops frameworks in Public Cloud, Private Cloud or Hybrid Environment

ML Ops Starter Kit Roadmap<a class="anchor" id="ml-ops-sk-roadmap">

Intended Roadmap of ML Ops Starter Kit. Can be changed based on priorities -

Contents of the Repository<a class="anchor" id="ml-ops-repo-contents">

How to get Started with this Starter Kit<a class="anchor" id="ml-ops-get-started">

Contributors<a class="anchor" id="ml-ops-contributors">

Please contact me for any questions.