Don’t Automate Until You’ve Perfected Things
Modern software engineering is all about automation. It should surprise no one that MLOps aims to take a lot of the automation principles from DevOps. However, should you always try to automate things?
That is a big question that many people have to contend with before putting ML in production. The reality is that automation isn’t always what you need when working with your ML pipelines. In fact, when it comes to MLOps, you might do better with less automation than more. You should not treat automation as a one-size-fits-all type solution that will help in everything. It should be noted on a case-by-case basis. However, you should also remember that the choice between automation and not isn’t a binary choice.
Automatic Doesn’t Always Mean Better
One thing to note is that automation is not always going to lead to better outcomes for your ML pipelines. The main issue is that certain industries are quite nuanced. The Machine learning lifecycle in those places will often be a lot different. For example, you might have some processes that must be done manually.
The drug discovery industry is one of those places where this is the truth. You need to do a lot of digging before you can start the process of building that ML automation pipeline. Because of those complexities, the practical thing to do is to wait a bit before embarking on doing machine learning in an automated way.
Learn the Process Before Automating It
Like with anything in life, it makes a lot of sense for you to learn how to do something before trying to automate it. It is the same thing in life when people tell you to learn to walk before you run. You need to have a lot of confidence in what you are doing before you try to start introducing automation. The beauty of MLOps is that it allows you to do that without spending extra time or effort breaking things down.
This field speeds up your journey from idea to a finished model, and you don’t have to do a lot of other things to get that to work correctly. The ML pipelines you have can be streamlined quite easily. Once you have gained some confidence that you are making the right decision, you are now ready to introduce automation to your machine learning lifecycle.
Not Everything Will Benefit from Automation
You can think of automation as another tool in your MLOps toolkit. The reality is that automation has to be more nuanced in your ML pipelines. The biggest benefits usually come from starting with the boring stuff, and from there, you can switch over automating the processes you have mastered. Ultimately, the most important thing is that your automation should focus on helping you bring your ML in production in the fastest time possible. You also want to ensure that you are making it as streamlined as possible to add to that. The faster you can iterate, the better off your machine learning models will be.