20.1 Recommendation with Machine Learning

We've had you working in ML Studio with "offline" CSV files so far. However, that's clearly not actual "machine learning" which requires that new data is constantly collected and used to retrain our models to improve our predictions automatically. This is a good time to both learn now to complete this machine learning loop and also get another chance to practice the recommendation system technique learned above. So, let's begin by getting a SQL Server database and website setup. Follow along with this video first:

If you've been able to setup you SQL Server AdventureWorks database and web app, and if you've been able to open, view, and edit the database through SQL Server Management Studio, and if you're able to open your web app in Visual Studio (via FTP) all like I did in the video above, then you're ready to move onto this next video. This video will show you how to create a recommender model that draws data from the SQL Server instance that you setup in Azure. Note that you'll need to understand Structured Query Language to copmlete this step. If needed, complete the Supplemental chapter in this book on "Reading data with SQL."

Next, let's turn that model into an API and use it on the website we created previously.

Now you have created a potential "feedback loop" where the same datasource that is used for collecting data is also used to train new predictive models. As a result, your predictions will influence consumer behavior (i.e. guide them to new products they will like more), that new consumer behavior (i.e. purchase decisions) is stored back into the same database, then that new data is used to continually re-train the Matchbox Recommender and continually improve and adapt the predictions. This is true "machine learning" because the predictions "learn" over time as old predictions influence new behavior. The final step is to automate the process of retraining so that no human intervention is needed.