uniform kernel

[ekinugurel]

Hi, relatively new GPyTorch user here. I am looking to implement a constant (uniform) kernel in the following form

I took the example sinc kernel shown in here and tried something similar using np.piecewise.

`class ConstantKernel(gpytorch.kernels.Kernel):
has_lengthscale = True

# this is the kernel function
def forward(self, x1, x2, **params):
    # apply lengthscale
    x1_ = x1.div(self.lengthscale)
    x2_ = x2.div(self.lengthscale)
    # calculate the distance between inputs
    diff = self.covar_dist(x1_, x2_, **params)
    # prevent divide by 0 errors
    diff.where(diff == 0, torch.as_tensor(1e-20))
    # return sinc(diff) = sin(diff) / diff
    return np.piecewise(diff, [diff <= -(self.lengthscale), 
                               abs(diff) <= self.lengthscale, 
                               diff >= self.lengthscale ], 
                        [0, 0.5*(diff+1), 1])`

But it doesn't seem to work. Can anyone advise on this approach? Sorry for the simple question.

[gpleiss]

What do you mean by "it doesn't seem to work"? Is it throwing errors? Returning incorrect results?

One thing of the top of my head - don't use np.piecewise. Numpy and torch functions don't always play well together.