homework 4, problem 1 hint

5 views
Skip to first unread message

percy...@gmail.com

unread,
Oct 24, 2007, 7:07:34 PM10/24/07
to CS281A: Statistical Learning Theory (Fall 2007)
Instead of the given hint, consider s_{ij} = s_{i(j-1)} and (z_{ij} or
(not x_{ij})).

Cohan Sujay Carlos

unread,
Oct 24, 2007, 10:30:36 PM10/24/07
to cs281a...@googlegroups.com
Percy,
 
Should the product on the RHS be over E_j to the power of x_j or is the Xi to be taken as a vector in the product?
 
Cohan
 

Percy Liang

unread,
Oct 24, 2007, 11:15:40 PM10/24/07
to cs281a...@googlegroups.com
In the first paragraph, we are talking about the distribution over a
binary vector x, in which case x_j is a scalar.

In the second paragraph, each data point x_i is a vector, so x_{ij}
is a scalar.

Ning

unread,
Oct 25, 2007, 9:15:16 PM10/25/07
to CS281A: Statistical Learning Theory (Fall 2007)
A silly question: what's the meaning of that inverse V sign in the
definition of s_ij?

Percy Liang

unread,
Oct 25, 2007, 10:26:57 PM10/25/07
to cs281a...@googlegroups.com
V means or.

Jeffrey Regier

unread,
Oct 28, 2007, 7:47:19 PM10/28/07
to cs281a...@googlegroups.com
Since s_{ij} is defined recursively, doesn't a base case also need to be specified? Would s_{i0} = 1?
 
Thanks.

 
On 10/24/07, percy...@gmail.com <percy...@gmail.com> wrote:

Percy Liang

unread,
Oct 29, 2007, 12:23:59 AM10/29/07
to cs281a...@googlegroups.com
Yes.

Ankit Jain

unread,
Nov 2, 2007, 1:24:29 AM11/2/07
to CS281A: Statistical Learning Theory (Fall 2007)
Where are the parenthesis in this equation? is it s_{ij} =
s_{i(j-1)} and [(z_{ij} or (not x_{ij}))]. OR s_{ij} = [s_{i(j-1)}

and (z_{ij}] or (not x_{ij})).

--ankit

On Oct 24, 4:07 pm, percyli...@gmail.com wrote:

Percy Liang

unread,
Nov 2, 2007, 2:43:37 AM11/2/07
to cs281a...@googlegroups.com
The former.

OzenHaman

unread,
Nov 3, 2007, 8:13:29 PM11/3/07
to CS281A: Statistical Learning Theory (Fall 2007)
The upside down V in the RHS means "AND" (an upright V is "OR").

So it's S_ij = z_i1 AND ... AND z_ij

-Omer

Reply all
Reply to author
Forward
0 new messages