how to generate stream with concept drift

[z. ahmadi]

Hi All,
I am using moa to generate a stream with drift but the problem is how can we control the drift?
to make it clear: my goal is to generate a dataset (for example hyperplane) with gradual drift, so I used ConceptDriftStream and selected the two generators to be hyperplane with the same number of features (set to be 2 for visualization) and then selected alpha and position of drift and width. I used a different random seed for each generators. the first time set the number of drifted attributes to be 0 and the result of visualization of stream showed that there was no drift in the stream. the second time I set the number of drifted attributes to be 2, then there are some drifts but not the one that I determined to occur at the specified position. lots of drifts without any control on them. so my question is now, is there any mechanism in moa to control drifts in the exact positions that we intend?
Thanks,
Zahra

[Albert Bifet]

Hyperplane has its own gradual drift (-k : The number of attributes
with drift, -t : Magnitude of the change for every example). If you
want to control the exact position of the drift, you can create a
meta-stream-generator in Java that implements this behaviour.. If you
want abrupt change you can use ConceptDriftStream.

Cheers, Albert

> --
> You received this message because you are subscribed to the Google Groups
> "MOA users" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to moa-users+...@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.

[kanu goel]

Hi All

I want to test a basic use case of concept drift for any dataset . I have tried a set a couple of values for the following parameters in the MOA GUI but unable to get any case of drift in the output values.

Please let me know the values to be inputted in the following parameters that are usually prompted in the GUI:

<<task?, learner?, base learner? ,drift detection method? [no of instances, warning Level, outcontrol Level] , dataset? , evaluator?>>

I have tried with the following set of values:

task:EvaluateAbruptDataset

learner:SingleClassifier

base Learner:Naive Bayes

DDM(30,2,3)resp

dataset:stagger

evaluator:BasicClassificationPerformanceEvaluator

In the output I am getting the value of Kappa Static is ranging from nearly 84 to 100%.PLease explain what column exactly denotes drift?

How exactly we evaluate that there is a drift occurance?

Correct me wherever wrong

Thanks a lot

Kanu

[Ahmad Idris Tambuwal]

Please! how can I simulate concept drift on existing data set (Cover Type) in UCI repository using MOA?

Thanks!

Ahmad