I am writing you with a some questions regarding the choice of the classifier options in the Trainable Weka Segmentation (TWS), as I have not been able to find any satisfactory information on the matter. I spent the last three weeks understanding all the essential aspects of the training,
I read several guides and tutorials on TWS but all of them fail to address the choice of specific classifier options, why or why not you should keep the default one (FastRandomForest) or opt for another one.
In this regard I have few unsolved questions:
What is the function of a classifier options? What I understood so far is that the default classifier (FastRandomForest) set the options in the segmentation settings panel, such as the choice of the training features, class names and number, membrane size and min/max sigma. Am I right?
What happens to the default classifier once I load my classifier model obtained by training the classifier (i.e., selecting different training features, modifying membrane size and sigma values, adding classes)? Is it deactivated and replaced? I cannot really understand this, because even after I loaded my model, opening the Segmentation settings panel, the FastRandomForest still appears in the classifier options as a choice.
What the sintax next to the classifier options name means? (e.g., for FastRandomForest is -I 200 -K 2 -S -841592)
Why I should choose another classifier options than the default one? How can I find the optimum one to my purpose?
Thank you in advance for all your answers and please, let me know if it’s not clear. As I am building a methodology, I need and want to understand everything in detail of what I am using, but so far what I found online is very far from being sufficient. Even the supposed classifier list page is unhelpful (http://weka.sourceforge.net/doc.dev/weka/classifiers/Classifier.html), very user UNfriendly to read and lacking in a clear explanation of what a classifier is and does.
Thanks a lot for your help!!