International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 11 Issue: 09 | Sep 2024
p-ISSN: 2395-0072
www.irjet.net
From Development to Deployment: Streamlining MLOps with Monoliths, Microservices, and Amazon SageMaker Karanbir Singh Senior Software Engineer, Salesforce, San Francisco, United States ---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In the era of AI, Machine learning models are
simple regression models to complex deep learning architectures.
integral to modern software applications. From spam detectors to self-driving cars, intelligent machine learning models are making their mark. However, the process of transitioning from model development to deployment poses significant challenges. This article aims to explore the model deployment process in detail and compare different deployment strategies such as Deploy as Monolithic, Deploy as Microservices, and Deploy using Amazon SageMaker. It also shed light on how Microservices and Amazon SageMaker can streamline and enhance Machine Learning Operations (MLOps). Additionally, it highlights relevant tools and practices that complement these approaches.
Training: The model is trained on the dataset to learn the underlying patterns and relationships. The goal is to optimize the model to generalize well to new, unseen data.
●
Model Evaluation: After training, the model is evaluated to ensure it is neither underfitting (failing to capture important patterns) nor overfitting (capturing noise in the data as if it were important).
●
Efficiency Computation: The model’s efficiency is computed in terms of its performance metrics, such as accuracy, precision, recall, or F1 score, depending on the use case.
●
Key Words: MLOps , Microservices, Monoliths, Amazon SageMaker, Kubernetes, Artificial Intelligence, Scalability, Model Development Lifecycle
2. Architectural Deployment
1. INTRODUCTION Developing a machine learning model is just the beginning of its lifecycle. To deliver value in a production environment, the model needs to be deployed efficiently, scaled to meet demand, and maintained over time. This process, commonly referred to as Machine Learning Operations (MLOps), encompasses the activities that ensure the model performs as expected when integrated into a broader application system.
Approaches
for
Model
Understanding how the model will be used in production as well as target audience is essential to guiding architectural choices for deployment
2.1. Monolithic Approach The monolithic approach involves deploying the entire system as a single, unified unit. This means that all components—whether they pertain to the user interface, business logic, or machine learning models—are tightly coupled and deployed together as a single application.
This article outlines the key stages of model development and explores the architectural choices for deploying models in a production environment, focusing on Kubernetes and Amazon SageMaker.
2.1.1. Example Application
1.1. Model Development: A Brief Overview
Use
Case:
Car
Dealership
Let’s consider a medium-sized car dealership that aims to provide personalized car recommendations to its customers based on their preferences. Since the target audience is relatively small, and the system does not require handling multiple versions of the application or models simultaneously, a monolithic architecture can be a practical choice.
The process of model development typically involves the following steps: Data Analysis: Understanding the data, cleaning it, and preparing it for training is the foundation of any machine learning model. This phase involves identifying patterns, relationships, and anomalies in the dataset.
●
Algorithm Selection: Depending on the use case, a suitable algorithm is chosen. This could range from
●
© 2024, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 16