LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Chapter 1 - The Learning Problem (http://book.caltech.edu/bookforum/forumdisplay.php?f=108)
-   -   How to choose features (http://book.caltech.edu/bookforum/showthread.php?t=4471)

netweavercn 01-31-2014 03:50 PM

How to choose features
 
I followed prof. Lin open course on coursera, which is great. does the book mentioned how to choose features, or I missed? (Give you have hundreds of features, you need them all? or you only need some of them)

magdon 02-05-2014 06:23 AM

Re: How to choose features
 
In general you should do feature selection and/or feature transformation to convert your set of many features into a set of a few useful features. Often, domain expertise is needed in this task, and Chapter 3 gives an example of constructing intensity and symmetry features for digit recognition, reducing the 256 "pixel features" to 2 useful features.

There are at least two good reasons to construct a few useful features. The first is that useful features usually simplify the problem and a linear as opposed to nonlinear separator may work. The second is that the hypothesis set will be simpler and so you will be able to generalize better to out-of-sample with fewer data.

Quote:

Originally Posted by netweavercn (Post 11636)
I followed prof. Lin open course on coursera, which is great. does the book mentioned how to choose features, or I missed? (Give you have hundreds of features, you need them all? or you only need some of them)


netweavercn 02-06-2014 02:08 AM

Re: How to choose features
 
Thanks.. makes sense.. I will check chapter 3


All times are GMT -7. The time now is 01:57 AM.

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.