"Balanced sampling"

frankie

A question. When I want to create a balanced training dataset using the "Sample" operator I get an error:
"Given index '4000' does not fit the mapped ExampleSet".

My data has two labels, "-1" and "1" and I try to take a 2000 sample from each of them.
The error comes during XValidation (not for example during preprocessing)

Can somebody help out, or guide me how these samples should/can be taken?
I realize it could be hard to pinpoint the source of the problem, but any ideas? The flow workds perfectly without the sampling operator.


-----------------------------------------------------

Exception: java.lang.RuntimeException
Message: Given index '4000' does not fit the mapped ExampleSet!
Stack trace:

  com.rapidminer.example.set.MappedExampleSet.getExample(MappedExampleSet.java:137)
  com.rapidminer.example.set.NonSpecialAttributesExampleSet.getExample(NonSpecialAttributesExampleSet.java:78)
  com.rapidminer.example.set.RemappedExampleSet.getExample(RemappedExampleSet.java:128)
  com.rapidminer.example.set.SplittedExampleSet.getExample(SplittedExampleSet.java:200)
  com.rapidminer.example.set.IndexBasedExampleSetReader.hasNext(IndexBasedExampleSetReader.java:62)
  com.rapidminer.example.set.AttributesExampleReader.hasNext(AttributesExampleReader.java:52)
  com.rapidminer.operator.learner.functions.kernel.evosvm.EvoSVMModel.performPrediction(EvoSVMModel.java:178)
  com.rapidminer.operator.learner.PredictionModel.apply(PredictionModel.java:76)
  com.rapidminer.operator.ModelApplier.doWork(ModelApplier.java:100)
  com.rapidminer.operator.Operator.execute(Operator.java:771)
  com.rapidminer.operator.execution.SimpleUnitExecutor.execute(SimpleUnitExecutor.java:51)
  com.rapidminer.operator.ExecutionUnit.execute(ExecutionUnit.java:709)
  com.rapidminer.operator.validation.ValidationChain.executeEvaluator(ValidationChain.java:211)
  com.rapidminer.operator.validation.ValidationChain.evaluate(ValidationChain.java:307)
  com.rapidminer.operator.validation.XValidation.performIteration(XValidation.java:143)
  com.rapidminer.operator.validation.XValidation.estimatePerformance(XValidation.java:133)
  com.rapidminer.operator.validation.ValidationChain.doWork(ValidationChain.java:261)
  com.rapidminer.operator.Operator.execute(Operator.java:771)
  com.rapidminer.operator.execution.SimpleUnitExecutor.execute(SimpleUnitExecutor.java:51)
  com.rapidminer.operator.ExecutionUnit.execute(ExecutionUnit.java:709)
  com.rapidminer.operator.OperatorChain.doWork(OperatorChain.java:368)
  com.rapidminer.operator.Operator.execute(Operator.java:771)
  com.rapidminer.Process.run(Process.java:899)
  com.rapidminer.Process.run(Process.java:795)
  com.rapidminer.Process.run(Process.java:790)
  com.rapidminer.Process.run(Process.java:780)
  com.rapidminer.gui.ProcessThread.run(ProcessThread.java:62)

Andrew2

Hello Frankie,

In the sample operator, set the sample parameter to relative and then set the sample ratio to 0.5. This will select half your examples.

regards

Andrew

frankie

What happens if the two groups contain 4000 and 3000 samples, respectively?
How will I then sample 2000 from each?


Thanks,
Frankie

TK

You can define absolute Values for each class within the sample operator

roman_bednarik

Hi, picking up on an old thread: how about if the size of the set is not known, e..g we don't know the absolute number of positive and the absolute number of negative examples? Is there a way to select a balanced subset?

MariusHelf

Hi, you can filter your data by label and then apply sampling operators on the filtered data sets and append them. I think http://rapid-i.com/rapidforum/index.php/topic,5706.0.html gives an example for that.

Best, Marius

abbasi_samira

Hello
How can I equal the number of classes (50 50) for two feature?

The class contains two values:

true:94

false:569

abbasi_samira

 

How can I balance the maximum amount of class attribute?

sgenzer

Hello @abbasi_samira - have you tried the “Balance Data” operator?

Scott

abbasi_samira

hi

I want to over sample balance data

How can I oversampling balance this?
Please explain the oversampling balance steps

please help me

thanks

abbasi_samira

hi

Yes I used the Balance Sample

read excel--->sample--->balance---->relative or absoulat

but This is the method undersampling balance

I need to oversampling balance

sgenzer

I would recommend going through the Sample operator tutorial (found inside the Sample help pane).

 

 

The Mannheim extension also has a Balance data operator.

Scott

 

kypexin

Hi @abbasi_samira

 

There's an extension called 'Operator toolbox' which now contains 'SMOTE upsampling' operator which you could use for oversampling the minority class.