As consultants, we’re always striving to expand the analytic capabilities we offer our clients; looking for new ways to deliver better value, faster. As data scientists, we understand the value of building these new competencies and skills – but we also just really like to geek out about cool, new technology. Digging into Azure ML gives us the opportunity to do both.
If you’re a fellow data scientist, a business leader looking for a game-changing technology solution or just a hard-core Microsoft enthusiast, here’s a high-level look at the benefits of automating the machine learning process with Azure ML.
Azure Machine Learning 101
The Azure Machine Learning Service is the next iteration of the Azure ML Studio platform. This is a drag and drop platform for creating, training and deploying machine learning experiments within the Azure cloud space.
The easiest way to understand the value of this new platform is by watching the webinar recording below. Recorded from a recent Microsoft Build event, this webinar does a great job of explaining the inherent issues with building complex models, including scalability, and shows where machine learning is headed.
Auto ML Webinar: https://bit.ly/2R5RP5u
Here’s a few spots in the recording of particular interest:
- Auto explanation starts at ~ 10:00
- Deployment options start at ~ 15:00
- Comparison of manual ML vs Auto ML runs from ~21:00 – 32:00 (this is interesting)
- Example of how Auto ML finds the best algorithm, starts at ~35:00 (this is VERY interesting)
- Walgreens Boots Alliance (Healthcare marketing use case) starts at ~41:00
- Mentions using a global set of hyperparameters for an ensemble of models at ~49:00
- 51:00 talks about value of Auto ML
Will Auto ML get it right every time?
Imagine an assembly line of Data Science projects where a team of Data Analysts work to aggregate and cleanse the data of customers looking to solve a problem. This data is then run through the Auto ML process, which quickly and efficiently determines the right algorithm and hyperparameters to use. Will Auto ML get it right every time?
Surprisingly, the experience of WBA at timestamp ~51:00, shows that in their scenarios, Auto ML produced better results than some of their hand-built models. But at a minimum, if it accelerates the tedious steps of model production, we can produce value for our clients much faster. The difficult problems can still be sent to one of our Data Scientists for custom coding and explanation if necessary, then plugged into a scalable cloud-based ML platform for easy deployment and maintenance.
As the ever-changing landscape of Data Science and Machine Learning continues to evolve, the demand for new services and offerings will increase as well. Thinking about where the data science puck will be in a few years is key to being able to guide our clients to their goals.