How can I compute/derive additional attributes?

Question

I have a dataset with two columns, A, and B.  For each record, i want to compute C = (B-A)/B.  Can I do that in RapidMiner transformations?  Can you direct me to the right one?

I also want to compute the overall C.  That is ((B1 + B2 + B3 ...) - (A1 + A2 + A3 ...)) / (B1 + B2 + B3 ...).  Can I compute this aggregate function?  How?  (I think this is a special case of a weighted average which I think I've seen reference to but haven't found the method for computing.)

Thanks.

MariusHelf · Answer

Hi, use Aggregate (not Create Aggregation) to create column-wise sum-aggregations. You will end up with a new example set, on which you can apply the formula (sum(A)-sum(B))/sum(A). Please note that Generate Attributes cannot handle attributes with parenthesis in their names, so you have to rename the aggregation attributes. Please have a look at the attached process. Happy Mining! ~Marius

ChrisNelson · Answer

When I pick "single" or "subset" instead of "all" for the filter type, a new control appears that allows me to pick attributes.  But then it appears that Generate Aggregation works across a single row.  What I need is something that works down the columns, perhaps this is creating new meta data?

Given:AB122341

I can use Generate Attribute to computer C = (B-A)/A for each row:ABC121.0230.541-0.75

But I need to calculate ((B1 + B2 + B3) - (A1 + A2 + A3)) / (A1 + A2 + A3):ABC121.0230.541-0.7576[glow=red,2,300]-0.14[/glow]

I guess I'm adding a new row, not a new column.  How do I do that?

ChrisNelson · Answer

In 5.2.008 on Linux, the Generate Aggregation has many more parameters in the help than in the form above it.  Notably missing is "attribute".  I see only:

* attribute name
 * attribute filter type
 * invert selection
 * include special attributes
 * aggregation function
 * keep all
 * ignore missings

Is this a bug or am I missing a step?