One of my work colleagues asked me what Azure ML Studio is. After I explained the tool to him, I realized I hadn’t used it for a few months. So, I figured I’d take it out for a test drive to see if there had been any changes since my last usage. Bottom line: no significant changes to Azure ML Studio since my last look at it. This is a good thing — a huge downside to Cloud based tools is that they can change far too often.
I did my usual Iris Dataset example. The image below summarizes ML Studio. Azure ML Studio is the ML.NET code library with a Web-based GUI interface. This makes ML Studio relatively easy to use but relatively difficult to document how to use it.
There is a lot going on in the demo. The first few blocks are the data pipeline where missing values, data normalization, and non-default data types are dealt with.
One thing that isn’t obvious when you’re new to ML Studio is the difference between Score Model and Evaluate Model. Score Model runs data through the model and generates a predicted class for each item and gives you the error but not whether the prediction is correct or not. The Evaluate Model component takes the predictions from Score Model and determines which predictions are correct and which are wrong, and computes classification accuracy.
Azure ML Studio is quite impressive. Because it’s not code-based, ML Studio is not suitable for all ML tasks. But for people who are new to machine learning, ML Studio makes an excellent learning tool.
Coincidentally, as I was writing this blog post, I got a request from the organizers of the Visual Studio Live conference to give a talk on Azure ML Studio. Excellent! Because I just verified that ML Studio doesn’t have any major changes, I am good to go. The conference runs March 1-6, 2020 in Las Vegas. See https://vslive.com.
Most people who have never been to a tech conference in Las Vegas imagine that it’s something like the image on the left. Sadly, the reality of tech conferences in Las Vegas is closer to the image on the right.