How to conduct a proper comparison between systems using SVM with RBF kernel?

Question

在比较使用带有RBF内核的SVM的系统时，我们应该对所有系统使用相同的C和伽玛（固定值，例如C = 10，gamma = 0.1）还是进行超参数转换，并为每个系统选择最佳的C和伽玛值系统？

Answer 1

As #sascha wrote, Different classifiers can train better with different parameters.

But it's OK. for different tasks you also have different requirement. If, for example you invented a new classification algorithm, and want to compare it with the state-of-the-art for a specific task, usually there will already be an optimized code for the specific task you are trying to learn, so you will only need to make small refinements to your own algorithm. On the other hand, if you have a new task, and just want to try out different classifiers, you will need to play with the both of them, until getting to some plateau. If there were big differences between the algorithms, take the one which gave you higher performance, and from than you can play only with it. If both of them gave you similar results, you have two options:

1. Make small adjustments to both of them until you get to a satisfy result
2. Try out other algorithms