Deploying a Machine Learning model to production

In a previous article I demonstrated how to build a User Interface for a machine Learning model using Django ML UI. In this article I will deploy the model to Heroku. You can find the Django ML UI package here:

What is Heroku?

Heroku is a Platform as a Service (PaaS) that allows developers to deploy, manage and scale applications. It supports a variety of programming languages including Ruby, Node.js and Python among others.

Deploying with GitHub

There are a few ways to deploy to Heroku. In this article we are going to configure the app on the local folder, push it to GitHub, pull it into Heroku, do a bit of config on that end and then deploy i.e. the Github method.

Step 1: Config on local machine.

First we're going to install Gunicorn, a Python Web Server Gateway Interface HTTP server and Whitenoise which will handle the static files:

pip install gunicorn

pip install whitenoise        

Next we are going to create a Procfile, this is a Heroku specific thing. Run this on the terminal.

touch Procfile        

Inside the Procfile add the following:

web: gunicorn <name_of_your_app>.wsgi --log-file -        

The name of my app is "config" so my Procfile will look like this:

web: gunicorn config.wsgi --log-file -        

Finally we're going to pip freeze the requirement file so that Heroku can run the requirements for our project.

pip freeze > requirements.txt        

Step 2: Push to GitHub

On the terminal, run the following to push your changes to Github:

git add .

git commit -m "commit"

git push        

Step 3: Heroku

Create a Heroku account, unfortunately they just discontinued their free tire but I think their Eco option is worth a try. On Heroko create a new app and give it a name (this will appear on the web address if you do not have a custom domain)

At this point you also need to add the host-name of your new app to the settings.py file:

ALLOWED_HOSTS = ['127.0.0.1', 'django-ml-ui-titanic.herokuapp.com']        

Go to settings (on Heroku) and add a build pack. This is a Python app so I will select Python:

Next go to Deploy, click on Connect to GitHub:

Then you will be prompted to grant Heroku access to your Github repos. Grant the access and connect to your repo. You should see something like this:

Now click Deploy Branch and in a moment your app should be ready. You should get confirmation as follows:

You should be able to now view the app and share it. The app deployed in this article looks like this:

Conclusion

This articles illustrates how to deploy a UI built with the Django-machine-learning-user-interface library to Heroku. Deploying machine learning models involves a lot more than getting the application to work and having a good UI does not guarantee project success but it sure beats excel files.