Linear Gaussian Process is not Linear

I want to implement a Gaussian Process model that is equivalent to Bayesian linear regression. This is my code. Assume that my data is 2D.

 with pm.Model() as self.model:
      l = pm.Gamma("l", alpha=2, beta=1)
      nu = pm.HalfCauchy("nu", beta=5)
      nu1 = pm.HalfCauchy("nu1", beta=5)
      cov = nu1 ** 2 + nu ** 2 *[1], l)

      gp =

      sigma = pm.HalfCauchy("sigma", beta=1)
      _y = gp.marginal_likelihood("y", X=X, y=y, noise=sigma)

      map_trace = [pm.find_MAP()]

But when I try to plot the function estimated by above Gaussian process model, I see some non-linear behaviors as well.


Above figure is generated by using the output from the linear GP model. As you can see, it is not linear (may be it is a combination of 3 linear functions). Is it usually behavior of linear GP?

Thanks in advance…


Below is the output using Linear regression. Which is the expected output.


gaussian process with linear kernel is linear between the points. Try with 1D so you see what is going on.

I think you want to do 2D linear regression, not GP?

1 Like

Thanks for the suggestion @ahartikainen . I will try that.

I actually want GP, because I’m trying to combine different kernels.

Try using different kernels with the same model, to see how the “jump” behaves across them? At the very least, I would think that would give insights into what is happening.