LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Homework 1 (http://book.caltech.edu/bookforum/forumdisplay.php?f=130)
-   -   Netflix movie recommendations (http://book.caltech.edu/bookforum/showthread.php?t=447)

mjbeeson 05-05-2012 07:07 AM

Netflix movie recommendations
 
According to lecture, each movie and each person is described by a vector of numbers; each number measures a "feature" that a movie can have (to some degree between 0 and 1), such as "comedy", or "dark", or "witty". ( I see these features mentioned by name in my Netflix recommendations.) Then the coefficients describing each movie and each person are adjusted by machine learning. But what I don't understand yet is where these features come from. Did a human invent them? or did a machine algorithm invent them?

Apparently in the banking example and the digit-recognition example, the key "features" were invented by humans, although, the last lecture hinted that the features in the banking example may have been machine-invented.
Feature-invention seems like a very key step in developing a particular application, so if there is a way to have the machine invent just the right features, that would be very interesting. I don't see anything in the textbook on this issue (there is an index entry for "feature selection", but it leads to p. 151 where the only mention of features is in the last few lines, about the digit-recognition example).

For example: Is there any algorithm that would take the raw data for the digit-recognition example and come up with the useful features? Also, a fingerprint-recognition example was briefly discussed, but the required features were not mentioned. What were they? and how were they identified? I am old enough to remember the time before computers could recognize fingerprints! Then it was considered a very difficult problem.

dudefromdayton 05-05-2012 10:51 AM

Re: Netflix movie recommendations
 
Someone please contradict me if I'm wrong, but if I remember correctly, fingerprint recognition is still a difficult problem. Or perhaps more accurately, fingerprints in general are not a reliable identification mechanism. They no longer have the standing in courts that they once did. My prints don't scan well on commercial equipment, and paper impressions from me have left the FBI stumped.

ladybird2012 05-07-2012 05:12 AM

Re: Netflix movie recommendations
 
Quote:

Originally Posted by mjbeeson (Post 1865)
But what I don't understand yet is where these features come from. Did a human invent them? or did a machine algorithm invent them?


I'm glad you asked this question. I have also been wondering about where the features come from for the netflix problem. The last lecture hinted that the algorithm comes up with these features (The prof made the point that we can't say what a feature meant in coming up with a rating for the netflix question). But then I'm not entirely sure I understand where the vectors u and v on slide 6 of lecture 10 come from...Unless "feature" on slide 6 is different from "feature" in the discussion on interpretation of hidden layers (slide 21).

loscan 05-10-2012 12:39 AM

Re: Netflix movie recommendations
 
Cool, I know netflix has a feature that recommends movies for you, but since I have 1 & 2 year old sons, all my recommendations are thomas the train, barney, and things like that.

jmknapp 05-10-2012 07:46 PM

Re: Netflix movie recommendations
 
Interesting. You have to wonder if the movie "features" are just nameless parameters if some of them would somehow end up as something intuitive like "comedy" or "lots of explosions" or "lead actor gets naked."

jiunjiunma@gmail.com 07-16-2012 09:41 PM

Re: Netflix movie recommendations
 
Yeah, I have the same question. Netflix's blog site has a good article describing their recommendation system, but they didn't say how they came out the features either.


All times are GMT -7. The time now is 09:36 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.