AWS re:Invent 2017 in Las Vegas attracted over 40,000 attendees. From the vendors to the conference itself, there were themes of predictive...
You are here
AWS Summit London - SageMaker for Machine Learning Beginners
AWS Summit London was held on May 10th at the ExCeL conference center in east London. This was my first time at an AWS conference, and I was blown away by the sheer scale of it. There had to be at least 10,000 attendees and every talk I attended was near capacity. I only had a general idea about the talks that I wanted to attend, but for some reason I was drawn to AWS's new product SageMaker. I had never heard of it until watching the keynote with Amazon.com CTO Dr. Werner Vogels. Many of my friends had studied Machine Learning (ML) at their university, but I didn't know anyone that has used it in industry. In the monitoring world, there are all sorts of interesting ideas around using ML for predictive analytics, so I wanted to know more about how other companies use ML and how SageMaker can be leveraged by companies wanting to get started in ML.
During Dr. Vogels' presentation, he was quick to point out that Amazon.com had been using ML for 20 years in their e-commerce business. It's easy to to see why this was important for them as they grew. On the product recommendation side, being able to intelligently upsell to customers must have made a huge impact on their sales. But, not many companies operate at the scale of Amazon. Can ML make a business impact for a small software startup or a traditional retail business?
What SageMaker was created for were those types of companies where ML isn't core to the business the way it is at Amazon. The hardest part of ML is the sheer effort involved in building, training and tuning your model so that it delivers accurate predictions for your business. Like many other of the services that AWS provides, they try to take away some of that complexity so that you can deploy quicker.
The first thing you need to do to build your model is connect it to your data. This is obviously much easier to do if you already have data stored on AWS, so this could be a blocker if you have it stored elsewhere. However, it doesn't take much effort to get some test data onto S3 so that the model can access it. Once you've got your data in place, you need to choose your algorithm. This is where a product like SageMaker really helps the beginner, because it already has algorithms preloaded with default settings to get you started. So, if we were to start using SageMaker in Opsview Monitor, we would most likely use the DeepAR algorithm, which uses large sets of time-series data to generate forecasts and predictions by analyzing patterns in that data.
Training is what separates ML from a simple input/output algorithm. The more data that is processed and the more time that passes, the better the model should get at predicting trends. Training is also the part that is the most computationally intensive, so again, this is where AWS tries to reduce the overhead. You only pay for what you use, so you can train your model until you're happy with the results. The tuning and tweaking that is necessary to get your algorithm configured correctly can also be done automatically by SageMaker.
This is where SageMaker provides real value to the business. Being able to deploy your ML solution on AWS allows businesses to quickly realize the benefit of their model and see how it performs in the real world. With features like one-click deployment, A/B testing, and hosting, you can be up and running with your first ML application without all of the overhead of deploying in on-premise infrastructure.
It can be intimidating getting started in the world of ML, but AWS has tried to lower the barrier of entry with the introduction of SageMaker. Microsoft and Google already have their own ML solutions, so it will be interesting to see how these three evolve over time. It's good that AWS has seen the value of ML for smaller businesses and has integrated it in with their other offerings, since they are the largest cloud service provider. I'm hoping this will allow more businesses to use ML in their day-to-day operations and benefit from the insight that it provides.
More like this
A detailed guide on how processing time series works to your advantage in Opsview 5.2.
New support for AWS services and regions helps IT teams monitor their cloud and on-premise environments within a single product.