Custom function for mean prior

[Jahnvi99]

How to define a negative quadratic function as prior for the mean?

[gpleiss]

class QuadraticMean(gpytorch.means.Mean):
    def __init__(self):
        # Define parameters for your quadratic
        self.a = -2.  # could be a learnable parameter

    def forward(self, x):
        return self.a * x.norm(dim=-1)

[Jahnvi99]

Thank you so much for providing me with the answer! But i am getting this error, using QuadraticMean() in the simple regression example of gyptorch

.

[gpleiss]

Can you include a reproducible example of your error?

[lukasfro]

For anyone that stumbles on this thread...

I got the same error message: 'QuadraticMean' object has no attribute 'training'.
What helped for me was to call the base class's __init__ method.
The following mean function works just fine for me:

class QuadraticMean(gpytorch.means.Mean):
    def __init__(self):        
        super(QuadraticMean, self).__init__()
        self.a = -2.0 

    def forward(self, x):
        return self.a * x.norm(dim=-1)

[gpleiss]

Oops, yes thanks for catching this @lukasfro !