I am trying to write code to build a growing SOM for mixed-type data. I came across a paper Growing Self-Organizing Map with cross insert for mixed-type data (http://www.sciencedirect.com/science/article/pii/S1568494612001731). It was very interesting and handled both categorical and numeric data in a unified way. However, my dataset has variables/attributes that can have multiple values (for example: attribute "interests" can have more than one value - movies, sports, and so on...). I got stuck at handling such attributes. Any inputs how to handle attributes with sets of values in a mixed-type dataset? References to material that talk about this problem would be greatly appreciated.
$\begingroup$
$\endgroup$
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
I have a couple of methods for such datasets. You might factorize the attributes into binary one-hot notation. If you have 3 possible values (car, truck, cycle) for an attribute, add new 3 dimensions to the row where each of them is the indicator of the one value. For instance for car value 100, truck 010, cycle 001.
Another way is to give unique discrete id to each possible value. car -> 1, truck -> 2, cycle -> 3, then use those integers instead of string names
As a caveat, better to use first method since SOMs are cannier to handle normalized data.
