Organizations are constantly seeking cost-effective solutions to reduce reliance on expensive third-party tools for both development and deployment of machine learning models. Recently, deploying a predictive machine learning model at our organization proved to be a significant challenge due to the high costs associated with infrastructure requirements.

Enter serverless computing, which offers a compelling solution for lightweight and on-demand ML inference. Platforms like AWS Lambda provide a particularly timely option given the rise in edge computing and machine learning use cases, allowing organizations to reduce excessive costs traditionally associated with ML deployment.

Deploying an ML Model on AWS Lambda

AWS Lambda is a preferred choice for ML model deployment due to its simplicity, automatic scalability, and cost-effectiveness. We only pay for the requests we make, making it a true pay-as-you-go service model.

Key advantages of using AWS Lambda include cost efficiency, scalability, and reduced infrastructure overhead. Serverless compute can potentially reduce infrastructure costs by up to 60% compared to maintaining dedicated prediction servers. Additionally, AWS Lambda automatically scales computational resources based on incoming prediction requests, eliminating the need for pre-provisioned server capacity.

However, it’s crucial to evaluate AWS Lambda’s limitations, including cold starts and resource constraints, to determine if it aligns with your specific ML deployment needs.

Approach #1: Deploying a Model Stored on Amazon S3

Deploying a ML model as a Python pickle file in an Amazon S3 bucket and using it through a Lambda API makes model deployment simple, scalable, and cost-effective. We set up AWS Lambda to load this model from S3 when needed, enabling quick predictions without requiring a dedicated server.

When someone calls the API connected to the Lambda function, the model is fetched, run, and returns predictions based on the input data. This serverless setup ensures high availability, scales automatically, and saves costs because you only pay when the API is used.

Creating a Lambda Layer

To deploy an ML model on AWS Lambda, we need to create a zip archive for the Lambda layer. A Lambda layer is a zip archive that contains libraries, a custom runtime, and other dependencies. We will demonstrate the creation of a Lambda layer using two Python libraries, Pandas and Scikit-learn, that are often used in ML models.

Below is the code for creating a Lambda layer zip archive, containing Pandas and Scikit-learn, using Docker. This approach allows us to package dependencies with our Lambda function, making deployment easier and more efficient.

In conclusion, deploying an ML model on AWS Lambda provides a cost-effective and scalable solution for organizations processing between 1,000 to 10,000 predictions daily. By leveraging serverless computing and AWS Lambda’s pay-as-you-go service model, organizations can reduce infrastructure costs by up to 60% compared to maintaining dedicated prediction servers.

Inspired by

• https://www.infoworld.com/article/4064124/how-to-deploy-machine-learning-models-with-aws-lambda.html

Sources

• python.org
• crayon.com