HomeChallengesZero down Time Machine Learning Architecture
learnbook poster
Zero down Time Machine Learning ArchitectureLast updated at Thu Dec 02 2021Skills

Learning Objective

After a successful attempt at the scenario, you will be able to

  1. Developing custom model in Sagemaker

  2. Understand different deployment endpoints in Sagemaker

  3. Deployment with API

Problem Statement

Caffepub is one of the leading coffee franchises in India. It has a presence in 18 states. It sells two kinds of coffee blends

  • Arabica
  • Robusta

Caffepub wants to launch a new promotional campaign in 3 poor-performing states. Using this campaign they want to increase profit and response rates. But promotional campaigns are always expensive and Caffehub wants to get the best returns on the investment. With this problem statement, they approached a reputed freelancer for a strategy. After careful analysis of data and previous market response models, the freelancer suggested that identification of Treatment responders (persuadable) is important for the success of the campaign.

Assuming you are the freelancer what would your approach be? What will your system architecture look like?

  • What modeling approach will you take? (Traditional uplift or a two-model approach. Will you use reinforcement learning?)

  • How will you finalize a better-performing model? (Will you base it on a QINI curve)

  • Will you use sagemaker in-built models or bring your own custom model?

  • Caffepub wants this to be a continuous process meaning the model developed will be used in different states and hence will be used multiple times with improvement and hence change to a better model should happen with zero downtime. Caffepub does not wish to have multiple instances to do this and is okay with shared memory and storage between containers. So which kind of inference endpoint will you choose? (Multi-Container or a Multi-Model)

  • How will you serve this model? (Will you use a micro service like architecture?)

Draw an architecture diagram of your approach before attempting solution

Help you can use

Wish to discuss the solution to this problem and solve more interesting problems?

Join us on Slack today!

Created with 💙 by
author avatar
Cloud Engineer,Mentorskool
mentorskool logo
Mentee Today, Mentor Tomorrow
No 206, A Block, Sonesta Silver Oak,Varthur, Bangalore 560066,Karnataka
Copyright - 2022 © Mentorskool - All rights reserved.