3

This is just for finding overfitting gap.
After initial research, I can only find method to draw learning curve using evaluation of test set. However, I could not evaluate on training set and over the two learning curves.

VividD
  • 636
  • 7
  • 18
modeller
  • 183
  • 5

1 Answers1

6

This is only possible with KnowledgeFlow. In WekaManual.pdf (which is included in Weka package) for version 3.7.12 there is an example in Chapter 7.4.2 "Plotting multiple ROC curves" with picture and step-by-step instructions. For other Weka versions it is the same, just find the appropriate chapter.

To give an impression on how it goes, I extracted the picture from the manual. It will draw two curves for two classifiers. For your question it is very similar. You use one classifier and then connect trainingSet to one ClassifierPerformanceEvaluator and testSet to another. enter image description here

lanenok
  • 1,476
  • 9
  • 9
  • Can you elaborate on this answer? I have been trying to carry out the modifications without success. In particular, how do you get the Knowledge Flow experimenter to retrain the classifier on increasing sizes of the training-set e.g 10%, 20%.... 90%, and then run the test-set on these trained models? Any help would be appreciated, preferably a revision to the diagram. – Orville Walker Feb 08 '22 at 02:44