I found that there may be bugs in Gaussian kernels in kernel SVM

[Weizhi-Zhao]

In this PPT page, when calculating the Gaussian kernel function, the square operation is performed after the two norms.
But the code, there is no square.

def _caculate_gauss_kernel(self, x, sv_x):
    n, dimension1 = sv_x.shape
    m, dimension2 = x.shape
    assert(dimension1 == dimension2)
    graph1 = np.repeat(np.expand_dims(x, axis=0), n, axis=0)
    graph2 = np.repeat(np.expand_dims(sv_x, axis=0), m, axis=0).transpose(1,0,2)


    kernel_value = np.exp(-self.gamma * np.linalg.norm(graph1 - graph2, axis=-1))

    return kernel_value

Is there any error? Or I am just wrongly understood your code.

[Daniel-xsy]

Sorry for the late response since I just notice the issue.

Yes, I think I forget to take the square at that time. Maybe you can try if taking square leads to better results.