- **Chapter 4 - Overfitting**
(*http://book.caltech.edu/bookforum/forumdisplay.php?f=111*)

- - **Problem 4.4: Interpretation**
(*http://book.caltech.edu/bookforum/showthread.php?t=4785*)

Problem 4.4: InterpretationCan someone confirm if this interpretation of "experiment" in part (d) is correct? I think we have nested loops:
for each choice of Q sub f, N, and sigma { Define the normalizing constant c sup 2 = E sub a, x (f sup 2). for each choice of coefficients {a sub q: q = 1,..., Q} from standard normal distributions { f(x) = sum from 1 to Q (a sub q L sub q (x)) / c for n = 1 to N y sub n = f(x sub n) + sigma epsilon sub n Find E sub out (g sub 2) and E sub out (g sub 10) } } The reason for all this detail is that I was unclear about what the conditional distribution p(y|x) might be. I think now that in each iteration of the second nested loop, we fix a choice of {a sub q}; given these coefficients, f becomes a deterministic function of x and the only randomness in y sub n is due to epsilon sub n. As a result, we have a different joint distribution P(x, y) for each choice of {a sub q}. We also have a different target function for each {a sub q}. The repeated experiments in (d), with fixed Q, N, and sigma, lead to one E sub out (g) for each set of coefficients {a sub q}. E sub out (g) is a function of the normal random vector (a sub 1, ..., a sub Q). The average of the out-of-sample errors is empirical mean of the distribution of E sub out. |

All times are GMT -7. The time now is 02:34 PM. |

Powered by vBulletin® Version 3.8.3

Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

The contents of this forum are to be used ONLY by readers of the Learning From Data book by Yaser S. Abu-Mostafa, Malik Magdon-Ismail, and Hsuan-Tien Lin, and participants in the Learning From Data MOOC by Yaser S. Abu-Mostafa. No part of these contents is to be communicated or made accessible to ANY other person or entity.