enter image description here In Figure 5.3, Pattern recognition and machine learning (Bishop), the author says he fitted 4 function: f(x) = x^2; f(x) = sin(x) ; f(x) = abs(x); f(x) = Heaviside(x), using 50 points chosen uniformly over the range [-1; 1]. And he did it with a 2-layer neural networks with 3 hidden units, and with "tanh" activation function. The problem is with the sine function: In the figure there are obviously a local minima and a local maxima, but we know that those 2 points are : x = pi/2 and x = -pi/2, and they are not in the range [-1; 1]. So how are we going to fit the sine function with the local minima (and maxima) if we don't even have any information about them ? Thank you very much


2 Answers 2


There's nothing about this in the 2011 errata to Bishop's PRML. If you believe that this is an error, you could contact the author.


As it appears in other scratches of code (see https://github.com/ctgk/PRML for Python), it's actually $sin(\pi x)$.

May you have a nice day!

  • 1
    $\begingroup$ Very good github page +1 $\endgroup$
    – Zhubarb
    Jul 9, 2018 at 10:59
  • $\begingroup$ But is this the official page? $\endgroup$ Jul 9, 2018 at 12:34
  • $\begingroup$ Not sure. Wouldn't think so...! But pretty good anyway. $\endgroup$
    – ailoher
    Jul 9, 2018 at 14:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.