'Poor fits for simple 2D Gaussian processes in `GPyTorch`

I'm having a lot of difficulty fitting a simple 2-dimensional GP out-of-the box using GPyTorch. As you can see below, the fit is very poor, and does not improve much with either swapping out the RBF kernel for something like a Matern. The optimization does appear to converge, but not on anything sensible.

class GPRegressionModel(gpytorch.models.ExactGP):
    def __init__(self, train_x, train_y, likelihood):
        super(GPRegressionModel, self).__init__(train_x, train_y, likelihood)

        self.mean_module = gpytorch.means.ConstantMean()
        self.covar_module = gpytorch.kernels.ScaleKernel(
                gpytorch.kernels.RBFKernel(ard_num_dims=2),
            )

    def forward(self, x):
        mean_x = self.mean_module(x)
        covar_x = self.covar_module(x)
        return gpytorch.distributions.MultivariateNormal(mean_x, covar_x)

Does anyone have good tutorial examples beyond the ones included in the docs?

pytorch gpytorch

Sources

This article follows the attribution requirements of Stack Overflow and is licensed under CC BY-SA 3.0.

Source: Stack Overflow

Solution	Source

'Poor fits for simple 2D Gaussian processes in `GPyTorch`

Sources

Related Questions