Docker for ML workloads in MLOps - Commands & Configuration

Introduction

Docker helps package your machine learning code, libraries, and environment into one container. This solves the problem of running ML models consistently on any computer or server without setup issues.

When you want to share your ML model with others and ensure it runs the same way on their machines.

When you need to deploy an ML model to a cloud server without worrying about missing dependencies.

When you want to test your ML code in a clean environment that matches production.

When you want to run multiple ML experiments with different library versions without conflicts.

When you want to automate ML training and deployment in a CI/CD pipeline.

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", "train.py"]

This Dockerfile starts from a lightweight Python 3.10 image.

It sets the working directory to /app inside the container.

It copies the requirements.txt file and installs Python packages needed for ML.

It copies all your ML code into the container.

Finally, it runs the training script train.py when the container starts.

Commands

This command builds a Docker image named 'ml-training-image' from the Dockerfile in the current folder. It packages your ML code and dependencies into one image.

Terminal

docker build -t ml-training-image .

Expected Output

Sending build context to Docker daemon  12.3MB

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-1.24.2

Removing intermediate container abc123def456

 ---> 567mno890pqr

Step 5/6 : COPY . ./

 ---> 234stu567vwx

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

 ---> Running in def789ghi012

Removing intermediate container def789ghi012

 ---> 890yz123abc

Successfully built 890yz123abc

Successfully tagged ml-training-image:latest

→

-t - Assigns a name and optionally a tag to the image

This command runs the container from the 'ml-training-image' image. The --rm flag removes the container after it finishes to keep your system clean.

Terminal

docker run --rm ml-training-image

Expected Output

Training started...

Epoch 1/10

Loss: 0.45

Epoch 2/10

Loss: 0.30

Training completed successfully.

→

--rm - Automatically removes the container when it exits

This command lists all Docker images on your system so you can verify your ML image was created.

Terminal

docker images

Expected Output

REPOSITORY          TAG       IMAGE ID       CREATED         SIZE

ml-training-image   latest    890yz123abc    2 minutes ago   150MB

Key Concept

If you remember nothing else from this pattern, remember: Docker packages your ML code and environment together so it runs the same everywhere.

Common Mistakes

Not including all required ML libraries in requirements.txt

The container will miss needed packages and your code will fail to run.

List all Python dependencies your ML code needs in requirements.txt before building the image.

Running docker run without the --rm flag during testing

Containers accumulate and use disk space if not removed after running.

Use --rm to automatically clean up containers after they finish.

Not copying the ML code files into the Docker image

The container will have no code to run, causing errors.

Use COPY commands in the Dockerfile to include your ML scripts and files.

Summary

Write a Dockerfile to specify the Python environment and ML dependencies.

Build a Docker image with your ML code and libraries using 'docker build'.

Run the ML training inside a container with 'docker run' to ensure consistency.

Use 'docker images' to check your built images and manage them.

Practice

1. What is the main benefit of using Docker for ML workloads?

easy

A. It provides a graphical interface for ML model training.

B. It automatically improves the accuracy of ML models.

C. It replaces the need for data preprocessing.

D. It packages the ML project with all dependencies to run anywhere.

Solution

1. Step 1: Understand Docker's role in ML

   Docker packages the ML project with all needed tools and code, ensuring consistency.

2. Step 2: Identify the main benefit

   This packaging allows the ML workload to run the same way on any machine without setup issues.

3. Final Answer:

   It packages the ML project with all dependencies to run anywhere. -> Option D

4. Quick Check:

   Docker ensures consistent ML environment = D [OK]

Hint: Docker bundles code and tools for consistent runs anywhere [OK]

Common Mistakes:

• Thinking Docker improves model accuracy
• Believing Docker replaces data preprocessing
• Assuming Docker provides a GUI for training

2. Which of the following is the correct syntax to start a Docker container named ml_container from an image called ml_image?

easy

A. docker start ml_image --name ml_container

B. docker create ml_image ml_container

C. docker run --name ml_container ml_image

D. docker build ml_container ml_image

Solution

1. Step 1: Recall Docker run command syntax

   The command to start a container with a name is: docker run --name [container_name] [image_name].

2. Step 2: Match the correct syntax

   docker run --name ml_container ml_image matches this syntax exactly, starting a container named ml_container from ml_image.

3. Final Answer:

   docker run --name ml_container ml_image -> Option C

4. Quick Check:

   docker run --name container image = B [OK]

Hint: Use 'docker run --name' to start named containers [OK]

Common Mistakes:

• Using docker start instead of docker run to create container
• Confusing docker build with running containers
• Wrong order of arguments in command

3. Given this Dockerfile snippet for an ML project:

FROM python:3.12-slim
WORKDIR /app
COPY requirements.txt ./
RUN pip install -r requirements.txt
COPY . ./
CMD ["python", "train.py"]

What happens when you run docker build -t ml_train . followed by docker run ml_train?

medium

A. The container only copies files but does not run train.py.

B. The container installs dependencies and runs train.py automatically.

C. The build command fails due to missing CMD syntax.

D. The container runs but does not install dependencies.

Solution

1. Step 1: Analyze Dockerfile commands

   The Dockerfile installs Python 3.12, sets /app as working directory, copies requirements.txt, installs dependencies, copies all files, then sets command to run train.py.

2. Step 2: Understand build and run behavior

   docker build creates an image with dependencies installed. docker run starts a container that runs train.py automatically as CMD is set.

3. Final Answer:

   The container installs dependencies and runs train.py automatically. -> Option B

4. Quick Check:

   Dockerfile CMD runs train.py after build and run = A [OK]

Hint: CMD runs train.py after build and run commands [OK]

Common Mistakes:

• Thinking CMD is ignored during run
• Assuming build fails without explicit entrypoint
• Believing dependencies install at run time

4. You wrote this Dockerfile for your ML project:

FROM python:3.12
COPY . /app
WORKDIR /app
RUN pip install -r requirements.txt
CMD python train.py

When building the image, you get an error: pip: command not found. What is the likely cause?

medium

A. The base image python:3.12 does not include pip by default.

B. The COPY command is incorrect and did not copy requirements.txt.

C. The CMD syntax is wrong and causes build failure.

D. The WORKDIR is set after COPY, causing path issues.

Solution

1. Step 1: Check base image contents

   Some python base images do not include pip by default, causing 'pip: command not found' error.

2. Step 2: Verify other commands

   COPY and WORKDIR are correct; CMD syntax is valid for shell form. The error points to missing pip in base image.

3. Final Answer:

   The base image python:3.12 does not include pip by default. -> Option A

4. Quick Check:

   Missing pip in base image causes error = A [OK]

Hint: Check if base image includes pip before installing packages [OK]

Common Mistakes:

• Blaming COPY command for pip error
• Thinking CMD syntax causes build error
• Ignoring base image contents

5. You want to optimize your Dockerfile for faster ML model training iterations by caching dependencies. Which change helps achieve this?

hard

A. Copy only requirements.txt and run pip install before copying the rest of the code.

B. Copy all files first, then run pip install to include all dependencies.

C. Run pip install after CMD to delay installation.

D. Use docker run to install dependencies each time the container starts.

Solution

1. Step 1: Understand Docker layer caching

   Docker caches layers. If requirements.txt changes, only pip install layer rebuilds, speeding up builds.

2. Step 2: Apply caching best practice

   Copying requirements.txt and installing dependencies before copying other code avoids reinstalling packages when code changes.

3. Final Answer:

   Copy only requirements.txt and run pip install before copying the rest of the code. -> Option A

4. Quick Check:

   Separate requirements.txt copy for caching = C [OK]

Hint: Copy requirements.txt first to cache pip install layer [OK]

Common Mistakes:

• Copying all files before pip install causing cache misses
• Running pip install after CMD which never executes during build
• Installing dependencies at container start wasting time