How to use SMOTE node in knime with integer dataset's columns?

KNIME Analytics Platform
1

Hi all,

I have to use the node "SMOTE" in the cross-validation metanode (I have put it after the X-partitioner and before the classifier, could it be right?). However, I have several errors when my dataset contains integer columns. I have tried putting a column rename before the cross-validation metanode, transforming the integer columns into double, but I got the following error:

 

The attribute 'Account length' has a different type in the test data (DoubleCell vs. Non-Native [interface org.knime.core.data.BoundedValue, interface org.knime.core.data.IntValue, interface org.knime.core.data.DataValue, interface org.knime.core.data.LongValue, interface org.knime.core.data.DoubleValue, interface org.knime.core.data.ComplexNumberValue, interface org.knime.core.data.FuzzyNumberValue, interface org.knime.core.data.FuzzyIntervalValue])

 

Could you help me?

 

By the moment, I'm normalizing the dataset before cross validation, but It's just a temporary solution, as you can see in the attached pictures (the first one is just illustrative, indeed the node outputs are red)

2

Hi,

thanks for reporting the problem. The weka predictor is indeed too restrictive. The problem is fixed and will be available with the next bugfix-release, probably this month. What KNIME version are you using? (seems to be lower than 3.0). I am not sure what other version, apart from 3.1, will be affected by this bugfix.

As another workaround for now you could also put a 'Round Double'-node after the SMOTE-node, because the SMOTE-node turns the former integer values into doubles and the 'Round Double'-node turns them back.

Best,

Martin

1 Like
3

The problem is fixed for 3.1.1 ...