I am studying Gaussian Processes for the first time on the second edition of “Bayesian Analysis with Python” by Osvaldo Martin.
In the example titled “Gaussian Process Regression” in this Jupyter Notebook https://github.com/PacktPublishing/Bayesian-Analysis-with-Python-Second-Edition/blob/master/Chapter07/07_Gaussian%20process.ipynb
everything seems to work fine as long as the max values of the sin function are close to 1 in magnitude.
If I replace y = np.random.normal(np.sin(x), 0.1) with y = np.random.normal(np.sin(x)*5, 0.1) here is what happens:
If I set the noise argument of gp.marginal_likelihood to one I get something that looks much better (unfortunately I can only upload one image)
Do you have an explanation for that?