Section 1.1: Understanding Microservice Architecture

The microservice architecture allows the decomposition of a large system into smaller, manageable components, enhancing scalability based on demand. Each component, known as a microservice, communicates over a network to ensure the larger system operates smoothly.

This approach simplifies the integration and maintenance of machine learning code within larger systems, enabling more frequent and reliable deployments to end-users.

In contrast, monolithic architectures pose challenges for scaling. If a single feature fails, the entire system is affected. Moreover, when demand for a feature spikes, additional resources must be allocated to the entire application instead of just the affected feature, leading to higher costs. The microservice model, however, allows for the independent scaling of components without disrupting the overall application.

Section 1.2: Virtualization vs Containerization

As mentioned earlier, virtual machines allow multiple operating systems to function on a single physical machine. This setup reduces the need for extensive hardware, lowers maintenance costs, and minimizes power and cooling requirements.

However, virtual machines depend on a hypervisor, which takes time to manage shared resources. Each microservice executed through a VM necessitates a replicated operating system, increasing resource demands.

In contrast, containers eliminate the need for a hypervisor, packaging the application and its dependencies into a single unit. This leads to greater efficiency, optimizing resource use and minimizing overhead.

The first video, "Introduction to Building ML Microservices: A Hands-On Approach With Examples From The Music Industry," offers insights into practical approaches to creating machine learning microservices, specifically focusing on real-world examples.

Section 1.3: Docker and Docker Compose

Docker has become synonymous with containers, serving as a management tool to package applications into portable images. These images can be shared and executed on any platform. Docker Compose is utilized to define and run multi-container applications.

With Docker, multiple applications can be containerized and run simultaneously on the same machine, sharing the operating system's services. This is particularly beneficial for deploying machine learning models across various global servers, ensuring scalability and functionality.

Docker also streamlines the management of interactions between microservices, simplifying reproduction, deployment, and updates.

Creating a Docker Image

Docker employs portable images to package applications. These images are constructed using a Dockerfile—a text document detailing the commands to assemble the image.

Each command in the Dockerfile creates a layer, which is cached for efficient reuse during subsequent builds. Below is a sample Dockerfile:

• FROM: Defines the base image (e.g., Python:3.9.4).
• WORKDIR: Sets the working directory for the application.
• RUN: Specifies commands to execute within the container, such as installing dependencies.
• EXPOSE: Indicates the port for application access.
• CMD: Defines the command to run when starting the container.

To build the Docker image, execute the docker build command in your command prompt. Before doing so, let’s introduce the application.

Section 2.1: Overview of the Fraud Detection Application

The data used in this project originates from a Kaggle competition in 2019, where participants were challenged to create a model for fraud detection based on historical customer transactions. My focus was not to develop the best model but to navigate through the entire process, excluding problem framing and data acquisition.

For more detailed insights about the application, check out the following resources:

• Automating Your Machine Learning Workflow with Scikit-learn Pipelines
• Testing Machine Learning Systems: Unit Tests
• Publishing a Packaged Machine Learning Model With Gemfury
• Deploying a Machine Learning Model as a Microservice Using FastAPI and Heroku

Section 2.2: Containerizing the ML Application

To build the image and start the container, you'll need two key commands: docker build and docker run.

Initiate the Docker image build with the command:

After building the image, confirm its creation by executing docker images.

To run the application from the image, use the docker run command, specifying the port, an environment variable, and the image name:

docker run -p 8001:8001 -e PORT=8001 fraud-detection-api

You can verify the application's operation by visiting localhost:8001.

Congratulations! You’ve successfully containerized a machine learning application.