LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Chapter 4 - Overfitting (http://book.caltech.edu/bookforum/forumdisplay.php?f=111)
-   -   Chosen #features for CV (http://book.caltech.edu/bookforum/showthread.php?t=4468)

nyc_ds 12-30-2013 08:01 PM

Chosen #features for CV
 
On page 152 (4.16b) E_cv is minimized at #features = 5 and 7. You say "the cross validation error is minimized between 5--7 feature dimensions; we take 6 feature dimensions..."

Why 6 and not 5, especially given the discussion about Occam's Razor that follows? 5 features would have same E_in, lower E_cv (and lower E_out but which you could not know in reality) and would be the simpler model. I know that it would make little practical difference but as a general principle shouldn't Occam's razor be favored?

yaser 12-31-2013 06:08 PM

Re: Chosen #features for CV
 
Quote:

Originally Posted by nyc_ds (Post 11626)
On page 152 (4.16b) E_cv is minimized at #features = 5 and 7. You say "the cross validation error is minimized between 5--7 feature dimensions; we take 6 feature dimensions..."

Why 6 and not 5, especially given the discussion about Occam's Razor that follows? 5 features would have same E_in, lower E_cv (and lower E_out but which you could not know in reality) and would be the simpler model. I know that it would make little practical difference but as a general principle shouldn't Occam's razor be favored?

Good point :). I guess we were trying to show how good the CV decision by itself will be, so we took a neutral view instead of resorting to additional tie-breakers like Occam's Razor.


All times are GMT -7. The time now is 02:55 AM.

Powered by vBulletin® Version 3.8.3
Copyright ©2000 - 2019, 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.