3

I am working through the book "Elements of Statistical Learning" and I do not understand why $Pr(dx,dy)$ is in equation 2.10

Background:

enter image description here

This is what I understand so far:

I am calculating the expectation value of the loss function $L(Y,f(x))$, thus:

$$ \begin{eqnarray} E[L(Y,f(x))] &=& \int L(Y,f(x))\,\, Pr(X,Y)\,dxdy \label{} \\ &=&\int[y-f(x)]^2\,Pr(x,y)\,dxdy \end{eqnarray} $$

The question: How does $Pr(dx,dy)$ appear in the above equation, e.g. what steps are taken to arrive at the equality below: $$ \begin{eqnarray} \int[y-f(x)]^2\,Pr(x,y)\,dxdy = \int[y-f(x)]^2\,Pr(dx,dy) \end{eqnarray} $$

Related posts I've read:

  1. Expected Error Prediction Derivation

    This post says $Pr(x,y)\,dxdy = f_2(x,y)dxdy$

    Is this just standard notation, i.e. that the derivative of a joint distribution function $Pr(x,y)dxdy$ is written as $Pr(dx,dy)$ or $f_2(x,y)dxdy$. In which case is $Pr(x,y) = f_2(x,y)$? So: $$ \begin{eqnarray} Pr(x,y)\,dxdy = Pr(dx,dy) \end{eqnarray} $$

    If this is not just notation, is there a mathematical derivation for $$ \begin{eqnarray} \int[y-f(x)]^2\,Pr(x,y)\,dxdy = \int[y-f(x)]^2\,Pr(dx,dy) \end{eqnarray} $$

Edit

Changed notation $f(x,y)$ to $f_2(x,y)$ to clarify it is distinct from $f(x)$

Community
  • 1
i_love_chocolate
  • 141
  • 5
  • 2
    First of all, you use the notation $f$ twice. Once for the statistical "model" $f(X)$ that should be close to $Y$, and once for the resulting joint density of the random variables $X$ and $Y$. That should be avoided. Secondly, assuming $X$ and $Y$ are continuous, it is indeed the case that $Pr(dx,dy)$ in the book really means $Pr(x\leqslant X – StijnDeVuyst Jan 17 '17 at 11:57
  • Yeah I know I used $f$ twice, the related post used $f(x,y)$ which really confused me. I replicated it in case it was intended. Thanks for the explanation regarding the notation, much appreciated! – i_love_chocolate Jan 17 '17 at 20:28
  • This issue is discussed in several posts on [prediction limits](http://stats.stackexchange.com/search?q=%22prediction+limit%22). In particular, see the extended discussions of specific prediction limits or intervals at http://stats.stackexchange.com/questions/4174, http://stats.stackexchange.com/questions/134380, and http://stats.stackexchange.com/questions/255570. – whuber Jan 17 '17 at 23:22

0 Answers0