Why containers make ML deployment portable in MLOps - Why It Works

Introduction

Deploying machine learning models can be tricky because different computers have different software and settings. Containers solve this by packaging the model and everything it needs into one neat box that works the same everywhere.

When you want to share your ML model with a teammate who uses a different computer setup

When you need to move your ML model from your laptop to a cloud server without breaking it

When you want to run your ML model on different machines without reinstalling all software

When you want to keep your ML model environment consistent during testing and production

When you want to avoid conflicts between different ML projects on the same machine

Config File - Dockerfile

Dockerfile

FROM python:3.10-slim
WORKDIR /app
COPY requirements.txt ./
RUN pip install --no-cache-dir -r requirements.txt
COPY . ./
CMD ["python", "serve_model.py"]

This Dockerfile creates a container image for an ML model.

FROM python:3.10-slim sets the base image with Python 3.10.

WORKDIR /app sets the working folder inside the container.

COPY requirements.txt ./ copies the file listing needed Python packages.

RUN pip install --no-cache-dir -r requirements.txt installs those packages.

COPY . ./ copies your ML model code into the container.

CMD ["python", "serve_model.py"] runs the model serving script when the container starts.

Commands

This command builds a container image named 'ml-model' with tag '1.0' using the Dockerfile in the current folder. It packages your ML model and its environment.

Terminal

docker build -t ml-model:1.0 .

Expected Output

Sending build context to Docker daemon  5.12MB

Step 1/6 : FROM python:3.10-slim

 ---> 123abc456def

Step 2/6 : WORKDIR /app

 ---> Using cache

 ---> 789def012abc

Step 3/6 : COPY requirements.txt ./

 ---> Using cache

 ---> 345ghi678jkl

Step 4/6 : RUN pip install --no-cache-dir -r requirements.txt

 ---> Running in abc123def456

Collecting numpy

Installing collected packages: numpy

Successfully installed numpy-1.23.1

Removing intermediate container abc123def456

 ---> 901mno234pqr

Step 5/6 : COPY . ./

 ---> 567stu890vwx

Step 6/6 : CMD ["python", "serve_model.py"]

 ---> Running in def789ghi012

Removing intermediate container def789ghi012

 ---> 345yz012abc34

Successfully built 345yz012abc34

Successfully tagged ml-model:1.0

→

-t - Assigns a name and tag to the image for easy reference

This command runs the container image 'ml-model:1.0' and maps port 5000 inside the container to port 5000 on your computer, so you can access the ML model service.

Terminal

docker run -p 5000:5000 ml-model:1.0

Expected Output

 * Serving Flask app 'serve_model'

 * Running on http://0.0.0.0:5000/ (Press CTRL+C to quit)

→

-p - Maps container port to host port for network access

This command sends a test request to the ML model running inside the container to get a prediction for input data [1,2,3].

Terminal

curl http://localhost:5000/predict -d '{"input": [1,2,3]}' -H 'Content-Type: application/json'

Expected Output

{"prediction": [0.5, 0.7, 0.2]}

→

-d - Sends data in the request body

→

-H - Sets the content type header to JSON

Key Concept

Containers bundle your ML model and all its software so it runs the same way on any computer.

Common Mistakes

Not including all required packages in requirements.txt

The container will miss needed software, causing the model to fail when running.

List every Python package your model needs in requirements.txt before building the container.

Forgetting to map ports with -p when running the container

You won't be able to access the ML model service from your computer.

Always use -p host_port:container_port to expose the service port.

Running the container without copying the model code inside

The container won't have your model files and will fail to start the service.

Use COPY commands in the Dockerfile to include your model code.

Summary

Use a Dockerfile to package your ML model and its environment into a container image.

Build the container image with 'docker build' to create a portable package.

Run the container with 'docker run' and map ports to access the model service anywhere.