Apply Model with less Attributes

t_liebe

Hey guys,

given a working model, is it possible to apply the model on an example set with less/different attributes? I have a lot of data on which I can build the model. However, I want to make a prediction on data where this data is not yet available.

Example:

ID	Happy (Label)	Text	Legal Age
1	true	Lorem ipsum dolor sit amet, …	true
2	false	Lorem ipsum dolor sit amet, …	false
3	false	Lorem ipsum dolor sit amet, …	true
4	true	Lorem ipsum dolor sit amet, …	true
5	false	Lorem ipsum dolor sit amet, …	false

As you can see, legal age correlates with the Label, but when I want to apply the model on data sets, I only have the texts. Is that possible ?

Thank you for your help.

MartinLiebig

Hi,
what should the model do in this case if it needs it? Some models can add missings here and evaluate like the Age would be missing.
BR,
Martin

t_liebe

The Problem is that the Attribute can't be evaluated easily.
Maybe I can explain my Goal with another example:

ID Blue Yellow Brown Green (Label) 1 1 1 0 1 2 0 1 0 0 3 1 0 0 0 4 1 1 1 0 5 0 1 1 0 6 1 0 1 0 ID Blue Yellow Green (Label) 1 1 1 50% 1; 50% 0 2 0 1 0 3 1 0 0

Although Attribute Brown is missing, you can still make a prediction based on the data set before. I know this might not be possible, but I thought it is worth a try to ask.

Kind regards,
Tobias

MartinLiebig

Hi,
don't you want to built a second model without the missing attribute in and then select during application which model to take?
BR,
Martin

Telcontar120

I agree, this seems like the case where a segmented scorecard based on underlying data availability would be the best solution.  This is a fairly typical setup in my experience.