CODEWITHSUNDEEP
pythondockermachine-learningflaskcontainers
GeorgeOfTheRF
GeorgeOfTheRF

Reputation: 8844

How to containerize a simple machine learning python application using docker?

The code in this repo shows how to create a flask web endpoint to predict probability of 'surviving the titanic disaster'. The trained model is serialized as pickle file using joblib which takes as input age, ticket_class, boarding_location and gender to make the prediction.

Training data - https://www.kaggle.com/c/titanic/data

Architecture of AWS Sagemaker Architecture of AWS Sagemaker https://docs.aws.amazon.com/sagemaker/latest/dg/how-it-works-hosting.html

The architecture in the above diagram looks like a good way to containerize and deploy ML application.

Question

  1. How do I use containerize for the ML + flask application in repo on my laptop? My objective is to deploy the container in production environment.
  2. How do I run the above container?

Upvotes: 1

Views: 1015

Answers (1)

ksbg
ksbg

Reputation: 3284

For simplicity, let's create our image based on Ubuntu.

Create a file Dockerfile in an empty directory with the following contents:

FROM ubuntu

# Install pip and git and clone repo into /app
RUN apt-get update && apt-get install --assume-yes --fix-missing python-pip git && git clone https://github.com/amirziai/sklearnflask.git /app

# Change WORKDIR
WORKDIR /app

# Install dependencies
RUN pip install -r requirements.txt

# Expose port and run the application when the container is started
EXPOSE 9999
ENTRYPOINT python main.py 9999

The image can then be built by running docker build -t <TAG> . from the directory containing the Dockerfile. You can then run it using docker run <TAG>.

I imagine you want to add your own sklearn models. There are different ways to do this, e.g. you could change your Dockerfile and use ADD to add files from your local file system to your image, or you could mount your model at the point of running the container.

Also, when actually using this container in production, you should consider a few more things, e.g. how to keep your container lightweight (e.g. by using alpine instead of ubuntu). You'll probably also want to upload your image to a hub, so you can deploy the container from there.

Hopefully this helps as a starting point.

Upvotes: 4

Related Questions