I wrote my first end-to-end functioning web application using Ruby on Rails in a class at MIT (6.170) in 2013. There were things that Rails automatically handled for me that I didn’t even realize were hard to do. Running
rails new just set up a completely functioning application. I never had to consider all of the components I would need to string together. Database migrations, routing, run and deploy scripts, tests, handling static assets, and more worked out of the box and the documentation clearly described how to build every part of my application. In fact, I assumed that writing web applications should always be this easy because I had never tried to write one from scratch. I was the beginner benefiting from my own ignorance that DHH talks about in The Rails Doctrine!
But beyond the productivity gains for experts, conventions also lower the barriers of entry for beginners. There are so many conventions in Rails that a beginner doesn’t even need to know about, but can just benefit from in ignorance. It’s possible to create great applications without knowing why everything is the way it is.
That’s not possible if your framework is merely a thick textbook and your new application a blank piece of paper. It takes immense effort to even figure out where and how to start. Half the battle of getting going is finding a thread to pull.
- DHH, The Rails Doctrine
A couple of years later, as a machine learning engineer at Slack, getting machine learning into production felt a lot more like "the framework as a thick textbook" and my application as "a blank piece of paper" that DHH talks about in the Rails Doctrine.
To make things even worse, try googling “how to learn machine learning”. The steps involved start looking like the curriculum required to obtain a PhD in Statistics, Math, and Computer Science.
The problems don’t end once you have successfully trained a model. You still have to figure out how to get your model into production. The code you wrote in your jupyter notebook needs to be translated into code that can be deployed. An entirely new job called “Machine Learning Engineer” was created just to solve this problem.
In the Rails Doctrine, there is a section on “Value Integrated Systems”. DHH says that Rails is “A whole system that addresses an entire problem.”
That’s a very broad scope, as we’ve discussed, but no broader than to be realistic to understand for a single person. Rails specifically seeks to equip generalist individuals to make these full systems. Its purpose is not to segregate specialists into small niches and then require whole teams of such in order to build anything of enduring value.
- DHH, The Rails Doctrine
One sentence in that section really stuck out to me: "Its [Rails'] purpose is not to segregate specialists into small niches and then require whole teams of such in order to build anything of enduring value". Today, this is exactly what companies are doing to get machine learning into production. They are required to assemble a team of specialists including Data Scientists, Machine Learning Engineers, Backend Engineers and Ops teams.
Tangram is an all-in-one automated machine learning framework that makes it easy to add machine learning to your applications. Predictions happens directly in your existing applications so there are no network requests and there is no need to set up a separate service to serve your models.
tangram trainto train a model from a CSV file on the command line.
tangram appto start a web application where you can learn more about your models and monitor them in production.
Tangram is a new project and there is a lot of work ahead. We’d love to get your feedback. Check out the project on GitHub, and let us know what you think! If you like what we are working on, give us a star!