Options

Can someone explain something about User k-NN ratings predictor?

Kholmes123Kholmes123 Member Posts: 1 Newbie
Hello. I am confused. I watched a training video on building a recommendation system using the User k-NN operator for ratings predictions.  In the video the instructor first used Set Role to not only set the label but he also typed in roles of item identification and user identification. These were not drop down menu items in  Set Role.   I can’t seem to find explanations of these two special roles in the documentation. 

My request:
Can someone please help me better understand these roles and if they are documented somewhere?  I don’t really feel like I know how to set it up correctly.  When I tried to run my first process on a case study it kept failing at the User K-NN with a mysterious and unexplained error.  I think I don’t have a good grasp. 

Answers

  • Options
    CKönigCKönig Administrator, Moderator, Employee, Member Posts: 70 RM Team Member
    Hi @Kholmes123,

    using the "Set Role" operator, you can assign special roles to certain attributes ("columns") in your data. Special attributes are generally treated differently from regular attributes, e.g. when training a Machine Learning Model, only regular attributes are used as input. For most operators, you need to check the "include special attributes" parameter, so special attributes also get changed. Special attributes are highlighted with colors in the "Data" panel of the Results View, and you can see the roles for special attributes in the "Statistics" panel.

    The behaviour of "Set Role" changed from RM version 9 to RM 10. Before, you could choose arbitrary role names, this is what most likely happened in that tutorial video. Since roles had to be unique, this was a good way to prevent problems with roles "overwriting" one another. Since version 10, we have a fixed set of available special roles. However, these can exist multiple times within an example set. 

    So if you want to exclude some attributes from being used as training ground truth, you can either remove them from the example set using Select Attributes, or you can keep them and use Set Roles and give them the role "metadata".

    See: Set Role | Altair Engineering Inc. Academy (rapidminer.com)
Sign In or Register to comment.