Next 'evaluate R' Bug: roles of true/false not detected correctly

fryasdffryasdf Member Posts: 7 Contributor II
edited November 2018 in Help
Hi.

I cant register to bugzilla due to another bug (dont you guys get paid for providing RapidMiner???).

Here is another bug concerning the 'execute R' operator:

When one wants to apply an R model to test data then one applies a prediction and one has to set the role to 'prediction' in order to be able to use the evaluation operators (performance, ...) from RapidMiner. However, RapidMiner does not see that 'true' means 'true', i.e. when doing a X-validation, then in the first round, 'true' means true but in the second round... RapidMiner is confused and thinks 'false' means true. Then I get the java exception from src/com/rapidminer/operator/performance/BinaryClassificationPerformance.java:

Cannot build average for different positive classes (true/false)

So what is missing here is to give the user access to another field in the mataData R object called 'orderOfValues' which should be a list with names being the values appearing in the table and the entry in the list being a string saying either 'TRUE' or 'FALSE' for binomial roles and an integer in the polynomial case.

The only solution I found so far is to enter a line 'false false' as the first line into the result data table. But am I sure then that this really gets interpreted as 'false' and not as the common 'true'? I dont know.


Tagged:

Answers

  • JEdwardJEdward RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 578 Unicorn
    Can you post up an XML of this so I can have a play?  I guess the Ripley Set in Samples would be good to use for demo.
    I don't want to be caught out when using similar.  I wonder if the Remap Binominals operator would help?  Have you tried that? 
Sign In or Register to comment.