LFD Book Forum

LFD Book Forum (http://book.caltech.edu/bookforum/index.php)
-   Homework 1 (http://book.caltech.edu/bookforum/forumdisplay.php?f=130)
-   -   *ANSWER* q6 (http://book.caltech.edu/bookforum/showthread.php?t=4182)

Moobb 04-09-2013 02:55 PM
*ANSWER* q6
 
I am confused on the answer for Q6. I can see that all choices have the same score, but I was assuming that not all of them would be valid hypothesis. I don't see how the opposite of XOR can agree with the 5 points in the dataset D. Isn't it required that the hypothesis would match those points??

yaser 04-09-2013 03:22 PM
Re: *ANSWER* q6
 
Quote:
Originally Posted by Moobb (Post 10298)
I am confused on the answer for Q6. I can see that all choices have the same score, but I was assuming that not all of them would be valid hypothesis. I don't see how the opposite of XOR can agree with the 5 points in the dataset D. Isn't it required that the hypothesis would match those points??
You are right that in learning we try to match the training points. The message of this problem is that regardless of whether you are doing something intelligent or otherwise, there is noting that can be learned outside the training sample in a deterministic sense. To make the message crisp, the problem considers some 'crazy' training schemes in the mix.

Elroch 04-09-2013 03:29 PM
Re: *ANSWER* q6
 
Quote:
Originally Posted by Moobb (Post 10298)
I am confused on the answer for Q6. I can see that all choices have the same score, but I was assuming that not all of them would be valid hypothesis. I don't see how the opposite of XOR can agree with the 5 points in the dataset D. Isn't it required that the hypothesis would match those points??
It is, and it can be presumed they do from the question. Given that the values of the function on D are already known, the definitions must only apply to the remaining 3 points.

[ apologies for cross-posting against the more authoritative answer above :) ]

yaser 04-09-2013 03:53 PM
Re: *ANSWER* q6
 
Quote:
Originally Posted by Elroch (Post 10300)
apologies for cross-posting against the more authoritative answer above :)
Thank you and all other participants who take the time to answer questions and make the discussion lively. Different points of view are always welcome. Just look at my signature. :)

Moobb 04-09-2013 04:38 PM
Re: *ANSWER* q6
 
Thank you! Must say that both answer 'features' were fundamental in order I could properly 'classify' the origin of my confusion about the question! :)

henry2015 09-02-2015 10:46 PM
Re: *ANSWER* q6
 
I just want to check my understanding at a different angle.

Is it true that the score is 1 x 3 + 3 x 2 + 3 x 1 + 1 x 0 for all?

And what idea behind this question is trying to pass to the students? Just trying not to miss the most important point in this question.

Thanks!

yaser 09-03-2015 12:42 AM
Re: *ANSWER* q6
 
Correct.

The idea is that, outside the training set, no "learning" is possible if we take a deterministic view. This is a formal version of the puzzle given at the end of Lecture 1.


All times are GMT -7. The time now is 08:45 AM.

Powered by vBulletin® Version 3.8.3
Copyright ©2000 - 2021, 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.