Scoring methods for neural network


  • I have a simple question about scoring methods in neural network algorithm.

    If I have attributes which have string values (for example male, and female), should I do normalization (male=1, female=2), or normalization could be done automatically in Visual Studio?

    I created model without normalization, and it works fine, and I am asking how huge mistake have I done in that way.

    Also, there are examples of DMX queries where user input string value instead of number. So, I suppose that model works fine without this process, or it is done automatically?

    Friday, January 10, 2014 8:23 PM


  • You don't need to map strings to numbers. SQL Server does this for you.

    There are many benefits of keeping your values as is and not doing what SQL Server does anyway. For example, when you look at the model viewer, you will see male and female instead of 1 and 2 which makes it easier to understand patterns discovered by the model.

    Tatyana Yakushev []

    Download Predixion Insight 3.0 - World class predictive platform for big data

    • Marked as answer by Mary44 Friday, January 10, 2014 9:42 PM
    Friday, January 10, 2014 9:21 PM

All replies