Fix issue 377

[idhamari]

New implementation to solve Issue 377: Enable definition of traning and validation set for privacy-preserving machine learning .

There is still a bug, probably related to this line:

        DataSubset subsetTrain = inputHandle.getSubset();

It seems the returned value is null when the handle comes from the GUI.

[prasser]

Please take a look at https://github.com/arx-deidentifier/arx/tree/feature-training-test

As mentioned, I merged your other PR into this branch and revised it. feature-training-test is the currently leading implementation of the feature, so if you want to further investigate this issue, please base your efforts on this branch. Will close this.

[idhamari]

Thanks for your feedback. I just tested the mentioned branch and I don't see the option for selecting kfold or subset for evaluation. I will add the UI part based on your recent updates.

[prasser]

The setting is there. It is contained in the project properties dialog.

[idhamari]

ok, I found the option "Use test and training set:" in Edit/Settings/Utility analysis.
It seems it stopped at this part

    if (config.isUseTrainingTestSet() && !inputHandle.isSubsetAvailable()) {
        throw new IllegalArgumentException("Training and test set can only be used with a subset");
    }

As the subset is not available despite I select random samples then did the anonymisation i.e.

          config.isUseTrainingTestSet()   : true
          inputHandle.isSubsetAvailable() : false

This error is similar to the one I got before in the previous implemntation!

[prasser]

Its not an error but expected behavior. You need to click the button to show the overall dataset.

[idhamari]

Thanks, everything works now after clicking the "Toggle sample view" button, maybe it should be activated automatically when "Use test and training set" is selected!