Altair RISE

A program to recognize and reward our most engaged community members

Nominate Yourself Now!

Performance (binominal) strange error!!!

mahtab3000

Hi,

I work with Performance(binominal) for getting "recall".

I use KDD'99 as dataset and it has a binary class named attack (values are: '0' , '1' ).

The running operation stoped with a strange error:

"The attribute 'prediction(label)' has 3 different values, must be 2 for calculation of 'recall'"

What does it mean?

Any idea's would be helpful.

Mahtab.T.

Find more posts tagged with

AI Studio

Errors

Performance

Accepted answers

All comments

Thomas_Ott

This error means that somewhere in your data set for the label, there is a third value. You can't see it but the computer finds it. Do this, go to the statistics Tabb, and see if the Data format is nominal not numerical. If that's the case then the nominal values are really strong values and not numerical values.

To solve the problem I would suggest using a parcel numbers operator to parse the numbers in the label column and then double check the statistics Tabb to see if they are now transformed from nominal to numerical.

Telcontar120

I have had a similar error before when a previous value is still present somehow in the metadata even after it has been removed in the data. In that case, you can also solve this by adding a "Remove Unused Value" operator to your process. This will ensure that only the values that are present in the data (in your case, first make sure your nominal label has only 2 values) don't affect any other operators via incorrect metadata.

Telcontar120

I understand you are still having this problem. Perhaps you can post your xml process and a data sample for further investigation? Here are a few additional ideas:

First make sure the "Remove Unused Values" has the "include special operators" option checked so it covers your label.
Next add a new operator to explicitly convert your label to binominal type (either "Numerical to Binominal" operator or "Nominal to Binominal") to make sure the metadata reflects that it is binominal.

Of course you should also check your data to make sure there really are only 2 values in the label as well.