This blog post was originally written for the company blog of my current employer, Sulava, and published on September 24th, 2020. As such the style and technical detail of this post do not reflect the posts written specifically for this blog. Regardless, I decided to publish these posts here as well for completeness’ sake.
Microsoft’s Ignite conference brought with it a ton of news as usual, and among those all was a long-awaited announcement of Azure’s Automated Machine Learning (ML) finally hitting general availability and leaving behind its over a year long preview state. Not only did the service reach GA, but Microsoft also announced that they will be merging the previously “Enterprise-grade” features of Azure Machine Learning with the cheaper Basic-tier, including Automated ML. This means that not only are these features now officially suitable for production use, but they are also very affordable to the customer: With Automated ML you will only pay for the resources you use, meaning that training new machine learning models can cost less than a cup of coffee does.
So, what is machine learning and why does this matter to me?
Of course, having a service reach GA and it becoming cheap to use is great and all, but with the myriad of different Azure services around we must ask ourselves why this one is something to be excited about? Machine learning itself is a data science discipline interested in finding out best possible answers to problems based on historical data. For example, you may have a list of patient records including details such as age, body-mass-index, blood pressure and whether the person is diabetic or not. Using this data, you could then train a machine learning model that can make predictions on whether someone else has diabetes based on their age, body-mass-index and blood pressure.
Similarly, you could train machine learning models to estimate daily sales, or the likelihood of a mechanical failure in a production machine, and many other things as long as you have the historical training data available. The movies that Netflix recommends you are a result of machine learning, too. While the results of machine learning are only predictions, these predictions can be made at rapid pace, faster than what any human is capable of. As such machine learning can produce valuable insights to large amounts of data that would be near impossible for humans to analyze, be it in real-time IoT systems or quarterly BI reporting. And even in daily decision-making machine learning can provide human decisionmakers with valuable second opinions that are less likely to hold any human biases.
But despite all the benefits that machine learning can bring us it has been rarely utilized, mostly due to the high barrier of entry: Producing accurate machine learning models has required a lot of manual work from highly skilled and sought-after data scientist specialists. Training the models has been an iterative process with lots of trial and error, and as such, the required investments needed for a machine learning project have been very high. Until now, that is.
Enter Automated Machine Learning
This is where Automated Machine Learning, Microsoft’s answer to making machine learning more affordable and available to all, comes in. It simplifies the process of training machine learning models by performing most of the iterative data science work for you, and turning projects that previously were 95 % data science and 5 % data engineering to more like 75 % data engineering and 25 % data science. Models can be trained and deployed to testing in a matter of hours, given good quality training data of course, and implementing proof-of-concept machine learning solutions with high quality models can be done quite handily. So, if you’ve thought that machine learning could be of use to you but were skeptical of the time and monetary investments required, worry no more.
Excited yet? Do get in touch and let us see how machine learning can help you!