Posts Tagged ‘Azure’

Thoughts after attending the Azure Machine Learning Meetup

November 26, 2014 Leave a comment

This is a long overdue post about the feedback after I attended the Azure ML meetup  hosted by Microsoft and Big Data Analytics user group. Overall, it was a great and informative session. Big thanks to Shashank Pawar and Rami Mukhtar.

The session stated with high level overview about what Azure ML’s capability. A few demos were given to showcase how easily you can leverage the ML to turn the predictive analytics into REST API that can be invoked easily. One of the fascinating demo Shashank showed is that you can call the ML web services on the fly from an Excel formula (CodePlex project – Azure Machine Learning Excel Add-In) while filling a table in Excel. Chris Webb has also blogged about using Power Query to do sentiment analysis for Facebook posts via Lexicon based API from ML.

One of the very interesting thing during the session is that no one, not even single one, asked about what algorithms Microsoft has been added into ML. The questions are pretty much focused on How R is integrated into the ML. People had asked about the performance, parallelism, available R packages, ability to upload custom R packages etc. It looks like most people who are interested in ML are actually considering using ML as the wrapper to operationalize R execution. And I’m not sure whether Microsoft thinks the same way. Will Microsoft treat the R execution as the first class citizen in the ML stack? ML offers a graphical interface to provide basic data exploring such as data profiling etc. I doubt how many people will actually use those features. I would imagine data scientist would just use their own “tool of the choice” and only use ML to operationalize the model. Personally, I would love to see the ML evolves to a predictive analytics automation engine with rich APIs for automation rather than a standalone predictive analysis packages.

A side note, one of the data source supported by ML is Power Query! This is a great idea and I’m really looking forward to seeing more demo of using PQ together with ML.