Project Botticelli

Boosted Decision Trees Visualisations

26 November 2015 · 1043 views

Azure Machine Learning

Microsoft have introduced quite a few updates to Azure Machine Learning in the last few weeks (and pretty much every month since the launch). The most recent one, and one greatly appreciated by this data miner/scientist, is the addition of Boosted Decision Tree visualisations. I am a big believer that a data model that we can understand and see is worth more than a blackbox approach. The excellent, interactive visualisations of the plain CART decision trees that we have in SQL Server Analysis Services are still some of the best available on the market. So I have been asking my friends at Microsoft to give us something similar in Azure ML. Some of my clients have a legal duty to explain how an algorithm, or their computer system, makes a decision. Not being able to understand the “why” is a blocker to the adoption of advanced analytics for these, and other organisations, not to mention the technology trust it prevents from being built.

The new Boosted Decision Tree visualiser in Azure ML

Of course, visualising a hundred trees created by the boosted algorithm is not easy, and even seeing all of them is not going to make things crystal clear, but it is a first step. I hope Microsoft improve this visualiser so it loads faster, please. But I also hope that they add more, also important visualisations, such as: cluster diagrams, profiles, discrimination charts and so on—many of which I explain in my data mining with SQL online course.

If you are new to all of this Azure ML stuff, have a look at my two introductory articles: