🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉
Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.
"[Solved] SVM parameter optimization of C and epsilon"
I have never been in the favorable situation to attend lectures on data mining. Thus, I lack the mathematical details on how they actually work. Nevertheless, I'd like to have at least an understanding of what is being influenced by the parameters.
To start into this topic I read some papers and searched the forum. Promising results were:
- "A User's Guide to Support Vector Machines" by Ben-Hur and Weston
In my case I have the "SVM (linear)" with a linear kernel. In the above mentioned post it is said that for this type of SVM only the parameter C has to be optimized. However, there are yet other parameters that can be set.
So I was wondering
- What effect do the parameters "C", "convergence epsilon" and "epsilon" have on the model at all?
- What are reasonable threshholds for these parameters for a (logarithmic grid) optimization?
I'd appreciate if someone could help with these questions or give a link for an "easy to understand" introduction to SVMs.