Lab 5: Containerizing with Docker

In this lab, you'll learn how to containerize a Flask application, train and deploy a machine learning model using Docker and Docker Compose. By the end of the lab, you'll have a Flask API running in a Docker container that can serve predictions using a machine learning model trained on the Iris dataset.

Clone the code from this repository.

Deliverables

• Setup Docker on your system
• Containerize training the ML Model
• Containerize the Flask App for inference
• Train and deploy the machine learning model using Docker

Deliverable 1 - Setup Docker

• Install Docker on your system
• Verify Docker installation

Follow the instructions for your operating system to install Docker. Then test your installation with:

docker run hello-world

Deliverable 2 - Containerize training the Machine Learning Model

Complete file named Dockerfile.train Fill in the TODO sections.

• Copy requirements.txt and install dependencies
• Copy train.py to the working directory
• Set the command to run train.py

Deliverable 3 - Containerize the Flask App for inference

Step 1: Implement the predict() Function

In the server.py file, there is a function called predict() where you need to:

• Load the trained machine learning model.
• Run inference using an input sent through a GET request.
• Return the prediction as the response.

Fill in the TODO sections.

Step 2: Complete Dockerfile.infer

Fill in the TODO sections.

• Set the working directory to /app
• Copy requirements.txt and install dependencies
• Copy server.py to the working directory
• Expose port 8080 or any other free port
• Set the command to run server.py

Deliverable 4 - Completing the Docker Compose

Docker Compose Overview

Docker Compose is a tool that allows you to define and manage multi-container Docker applications. It uses a docker-compose.yml file to configure the application's services, networks, and volumes. In this lab, you'll use Docker Compose to set up two services:

• Training Service: This service will train a machine learning model using the Iris dataset.
• Inference Service: Once the model is trained, the inference service will load the trained model and serve predictions via a Flask API.

Fill in the TODO sections:

• Set the Dockerfile.train for the training service and Dockerfile.infer for the inference service.
• Use a shared volume (model_storage) to store the trained model between the services.
• Expose port 8080 (or nay other port) for the Flask app in the inference service.
• Ensure the inference service starts only after the training service completes (depends_on).
• Declare the model_storage volume at the end for both services to access the trained model.

Running the Entire Setup

• Build and run services with Docker Compose
• Verify both services are running correctly

Use the following command to start your services:

docker-compose up --build

Testing the Prediction Endpoint

Calling a GET Request

For this assignment, run the following CURL command to test your flask setup.

curl --location --request GET 'localhost:8080/predict' \
--header 'Content-Type: application/json' \
--data '{
    "input": [6.3, 3.3, 6 , 2.5]
}'

Alternatively, you can test this on Postman as well - ensure that your body is JSON with the following data

{
    "input": [6.3, 3.3, 6 , 2.5]
}

Additional resources

1. Docker For Beginners
2. Build and Deploy Flask Applications with Docker
3. Models on Docker

Troubleshooting

If you encounter issues:

• Check Docker daemon status
• Verify port availability
• Review service logs with docker-compose logs
• Ensure the training service completes before the inference service starts