We have TONS of videos to help you learn RapidMiner - from beginner to expert. Click to watch!
We're competing as Team "RapidMiners" in DrivenData's latest competition: "Pover-T". Join now!
Read about how our community works. Meet other newbies. Get your questions answered fast!
I dont understand why one would want to perform SVM regression with C=0. All papers that i read in that topic talk about C > 0, because otherwise points do not get punished at all if they lie outside the tube, which means the position of the tube does not matter at all and the result should be very bad. Is the default value for C really 0 or is it 0.000001 or something like that? I also realized you can enter C=-1. What is this supposed to mean? And how can I set the value to infinity to get hard boundaries?
Thanks in advance for every reply!
Solved! Go to Solution.
Thanks for asking - I just realized that the docs are not making a good job explaining this. Anyway, a value of 0 indicates that we use a heuristic described in this post here to determine a value for C based on the provided data:
Unfortunately, I still do not remember exactly where we found this
Negative values are indeed not possible (nor do they make sense), not sure why the UI allows -1...
Hope this helps,
Thanks for your fast and helpful answer!
Is there a way to see the values that are used at the end? At Kernel Model I only get number of Support Vectors and the weights.
And another question: Is epsilon=0 used or is there another heuristic used then?
Good point, unfortunately the chosen value for C is not shown at the end. Might be a good thing to add at least for the text output (Description)...
No heuristic is used for epsilon, so 0 really means 0 here.
Hope this helps,