Packaging and containerization

End-to-End Machine Learning

Joshua Stapleton

Machine Learning Engineer

Deployment and containerization

Deployment

Packing model + dependencies into units
For running in different environment
De-facto framework for containerization and deployment is Docker

Deployment phase of machine learning lifecycle

Docker

Platform for simplifying development using containers

Containers:

Package application into standalone assets
Containers designed as platform-agnostic
Get Docker

DataCamp Docker Course

Docker logo

Docker usage part 1

Dockerfile: instructions for building container

# Use an official Python runtime as a parent image
FROM Python:3.7


# Set the working directory in the container to /app
WORKDIR /ML_pipeline


# Copy the current directory contents into the container at /app
ADD . /ML_pipeline


# Install any needed packages specified in requirements.txt
RUN pip install --no-cache-dir -r requirements.txt

Docker usage part 2

# ... continued
# Make port 80 available to the world outside this container
EXPOSE 80

# Define environment variable
ENV NAME World

# Run app.py when the container launches
CMD ["Python", "ML_pipeline.py"]

Build the defined image:

docker build -t heart_disease_model .

Tagging containers

Tagging:

docker tag heart_disease_model:latest heart_disease_model:1.0

Makes images / containers easier to identify and manage.
Helps in maintaining a detailed and robust model registry.
After tagging, we are ready to deploy!

Deployment visualized as an arrow

Best practices

While Docker makes packaging models easy...

Be security-minded
Don't include sensitive data
Use trusted images (from verified developers)

If you application does have sensitive information...

Use environment variables
Eg: for connection strings/passwords

Security lock

Let's practice!

End-to-End Machine Learning