I tried to reproduce Figure 8.14 from page 255 of your book, but I simply don't get the same result for the Bayes MC solution. The problem is that the prediction intervals outside the training range do not increase as shown in the book. They appear to increase more or less linearly (see appended plot).
It works fine with the weights loaded from github, but the code in the notebook cannot reproduce it. I executed the whole github notebook (chapter_08/nb_ch08_03.ipynb) using 2048 samples and the notebook cell with the corresponding fit function, but it seems that the github notebook was not the version used to obtain the results shown in the book.
Could you please provide the code that is able to reproduce the results from the book? Otherwise it is a somewhat sobering conclusion as you argue that the Bayesian models should be able to capture the uncertainty in unseen data regions.
I tried to reproduce Figure 8.14 from page 255 of your book, but I simply don't get the same result for the Bayes MC solution. The problem is that the prediction intervals outside the training range do not increase as shown in the book. They appear to increase more or less linearly (see appended plot).
It works fine with the weights loaded from github, but the code in the notebook cannot reproduce it. I executed the whole github notebook (chapter_08/nb_ch08_03.ipynb) using 2048 samples and the notebook cell with the corresponding fit function, but it seems that the github notebook was not the version used to obtain the results shown in the book.
Could you please provide the code that is able to reproduce the results from the book? Otherwise it is a somewhat sobering conclusion as you argue that the Bayesian models should be able to capture the uncertainty in unseen data regions.