http://backend.userland.com/rss092CSE Forums
https://cgi.cse.unsw.edu.au:443/~forums/support/
Working together at CSECSE Forums Digest Robot CSE Forums Digest Robot Mon, 27 Apr 2015 04:03:07 GMTCOMP9318 Assignments :: RE: Clarification of ChooseRule(s)
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73594#73594
Sun, 26 Apr 2015 07:01:58 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=3812" target="_blank">ajcu764</a><br />
Posted: Sun Apr 26, 2015 5:01 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Thanks so much guys. I understand it a lot better now.
<br />
<br />
Turns out I was dividing the dot product by the magnitude to get the projection. I just needed the dot product alone.
</span><br />
COMP3141 :: Exercise 5 - Compile Time Error
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73593#73593
Sun, 26 Apr 2015 04:47:59 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=4766" target="_blank">jjla809</a><br />
Subject: Exercise 5 - Compile Time Error<br />
Posted: Sun Apr 26, 2015 2:47 pm (GMT 10)<br />
Topic Replies: 0<br /><br />
<span class="postbody">Can someone just remind me if an error message like
<br />
"Couldn't match type ‘'Empty’ with ‘'NonEmpty’" is a compile time error?
<br />
<br />
What i'm trying to wrap my head around is if in ghci, i go
<br />
:l Ex05.hs
<br />
and it says it compiles fine but then when i run
<br />
top Leaf
<br />
and it comes up with the above error message, is that a compile time error or a run time error?
<br />
<br />
Thanks
</span><br />
COMP9318 Assignments :: RE: RP-tree algorithm
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73592#73592
Sun, 26 Apr 2015 02:00:58 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=102" target="_blank">weiw</a><br />
Posted: Sun Apr 26, 2015 12:00 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">You are right. My pseudocode is not strict - so does the pseudocode in the original paper!
<br />
<br />
Thanks for pointing it out.
<br />
<br />
Wei
<br />
<br />
</span><table width="90%" cellspacing="1" cellpadding="3" border="0" align="center"><tr> <td><span class="genmed"><b>jlta135 wrote:</b></span></td> </tr> <tr> <td class="quote">Okay, as I understand the rule split points on a projection plane by the hyperplane i.e. the median. So Rule(x) means project vector x to the projection plane (to convert it to scalar) and then compare it with median (which is also scalar). I think this is what it means.
<br />
<br />
So in pseudo code Rule(x) can be written more like (<x,u> <= med),
<br />
where the inner product means projecting x to projection plane u. Correct me if this is not the case.</td> </tr></table><span class="postbody">
</span><br />
COMP9318 Assignments :: RE: Clarification of ChooseRule(s)
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73591#73591
Sun, 26 Apr 2015 01:58:41 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=102" target="_blank">weiw</a><br />
Subject: Re: Clarification of ChooseRule(s)<br />
Posted: Sun Apr 26, 2015 11:58 am (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">I saw quite a few similar questions. I think the basic problem is that you need to refresh the notion of (scalar) projection of a vector onto another vector, which is computed by dot product (with normalization if needed). Anyway, if you draw a figure (or use the one on the wikipedia page), you will easily get the intuitive meaning of "projection" - the same way an object has a shadow under sunlight.
<br />
<br />
c.f., <a href="https://en.wikipedia.org/wiki/Vector_projection" target="_blank">https://en.wikipedia.org/wiki/Vector_projection</a> (we use the "scalar" projection) here.
<br />
<br />
<br />
<br />
<br />
</span><table width="90%" cellspacing="1" cellpadding="3" border="0" align="center"><tr> <td><span class="genmed"><b>ajcu764 wrote:</b></span></td> </tr> <tr> <td class="quote">Hi,
<br />
<br />
I was hoping someone could please clarify how Section 3.3 in the spec aligns with Slide 28 or at least what's the process behind ChooseRule(). I'm a bit confused as I'm not generating the right RP-Trees.
<br />
<br />
- U Vector. My understanding is that you get 3 numbers from guassian_rand.txt and create the vector then normalize it by dividing each value in U by U's magnitude.
<br />
<br />
- What is the set of n sorted numbers? I understand this as the set of <x, u> for every x in S according to the slide. Or is this actually the set of projections of x on U for every x in S?
<br />
<br />
- Is the projection of a point P, its projection onto U and calculated as <p, u> divided by the magnitude of P?
<br />
<br />
Any help or advice is greatly appreciated.
<br />
<br />
Thanks,
<br />
Aaron</td> </tr></table><span class="postbody">
</span><br />
COMP9318 Assignments :: RE: Assignment1 Q4
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73590#73590
Sun, 26 Apr 2015 01:53:55 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=102" target="_blank">weiw</a><br />
Subject: Re: Assignment1 Q4<br />
Posted: Sun Apr 26, 2015 11:53 am (GMT 10)<br />
Topic Replies: 1<br /><br />
<span class="postbody">d = #dimensions of the vector = 20. i.e., each vector consists of a sequence of 20 scalar values.
<br />
<br />
d has nothing to do with r and b (think in this way, if we do not use LSH, we do not need r and b, but d exists as long as the data is given).
<br />
<br />
</span><table width="90%" cellspacing="1" cellpadding="3" border="0" align="center"><tr> <td><span class="genmed"><b>SOLKI wrote:</b></span></td> </tr> <tr> <td class="quote">Hi,
<br />
Let me be quick.
<br />
In the Spec, "The dimensionality of the binary vectors are d = 20". Is this 20 the result of AND-then-OR construction or not? I'm confused with this for 1 day and if d is the dimensions of the original matrix, I mean the matrix before constructions, then the question 4.2 can not make any sense with r = 10 and b = 20. The product of r and b is 200, much larger than the dimension d which is only 20.
<br />
Hope someone could help, thank you!
<br />
<br />
Cheers!
<br />
<br />
Sol</td> </tr></table><span class="postbody">
</span><br />
COMP9318 Assignments :: RE: RP-tree algorithm
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73589#73589
Sat, 25 Apr 2015 12:30:23 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=2967" target="_blank">jlta135</a><br />
Posted: Sat Apr 25, 2015 10:30 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Okay, as I understand the rule split points on a projection plane by the hyperplane i.e. the median. So Rule(x) means project vector x to the projection plane (to convert it to scalar) and then compare it with median (which is also scalar). I think this is what it means.
<br />
<br />
So in pseudo code Rule(x) can be written more like (<x,u> <= med),
<br />
where the inner product means projecting x to projection plane u. Correct me if this is not the case.
</span><br />
COMP9318 Assignments :: RE: RP-tree algorithm
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73588#73588
Sat, 25 Apr 2015 11:08:06 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=3645" target="_blank">ldawes</a><br />
Posted: Sat Apr 25, 2015 9:08 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Each rule must contain both a vector and a scalar. Between them they define the splitting hyperplane.
</span><br />
COMP9318 Assignments :: RP-tree algorithm
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73587#73587
Sat, 25 Apr 2015 10:28:47 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=2967" target="_blank">jlta135</a><br />
Subject: RP-tree algorithm<br />
Posted: Sat Apr 25, 2015 8:28 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Hi, I'm not sure whether I'm understanding the RP-tree pseudo-code properly.
<br />
<br />
The ChooseRule(S) method returns (Rule, U, boundary). Rule is defined as λy. (y ≤ med). This takes a parameter y and returns a boolean (y <= med). So y is a scalar value here.
<br />
<br />
Yet in the MakeTree(S) method,
<br />
Node.LeftChild <- MakeTree({x ∈ S | Rule(x) = TRUE })
<br />
x is a vector within set S. How can Rule(x) take in vector x when it takes a scalar value?
<br />
<br />
Any clarification on this is much appreciated. Thanks
</span><br />
COMP9318 Assignments :: RE: Assignment 1 Question 4.4
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73586#73586
Sat, 25 Apr 2015 02:57:21 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=3645" target="_blank">ldawes</a><br />
Posted: Sat Apr 25, 2015 12:57 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">My question has been clarified in the latest release of the spec. I think d=20 is the dimensionality of the search space, that is, the number of features your data and query vectors have.
</span><br />
COMP9318 Assignments :: RE: Clarification of ChooseRule(s)
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73585#73585
Sat, 25 Apr 2015 02:15:16 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=5071" target="_blank">z5039854</a><br />
Posted: Sat Apr 25, 2015 12:15 pm (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">U vector - my understanding is that for each internal node you create your vector from the gaussian file - take as many numbers as required.
<br />
<br />
I found n sorted numbers implied that for each element x in S, find <x,u> and then find the median of all those dot products. So, the set of the projections you create.
<br />
<br />
I took projection to be <x,u>. That's the x vector you have dot producted with the unit vector u (the initial u from the gaussian file, then normalised).
<br />
<br />
Hari
</span><br />
COMP9318 Assignments :: Clarification of ChooseRule(s)
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73584#73584
Fri, 24 Apr 2015 15:48:43 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=3812" target="_blank">ajcu764</a><br />
Subject: Clarification of ChooseRule(s)<br />
Posted: Sat Apr 25, 2015 1:48 am (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Hi,
<br />
<br />
I was hoping someone could please clarify how Section 3.3 in the spec aligns with Slide 28 or at least what's the process behind ChooseRule(). I'm a bit confused as I'm not generating the right RP-Trees.
<br />
<br />
- U Vector. My understanding is that you get 3 numbers from guassian_rand.txt and create the vector then normalize it by dividing each value in U by U's magnitude.
<br />
<br />
- What is the set of n sorted numbers? I understand this as the set of <x, u> for every x in S according to the slide. Or is this actually the set of projections of x on U for every x in S?
<br />
<br />
- Is the projection of a point P, its projection onto U and calculated as <p, u> divided by the magnitude of P?
<br />
<br />
Any help or advice is greatly appreciated.
<br />
<br />
Thanks,
<br />
Aaron
</span><br />
COMP9318 Assignments :: RE: About FindNN Algorithm, i have a question
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73583#73583
Fri, 24 Apr 2015 14:18:45 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=5087" target="_blank">SOLKI</a><br />
Posted: Sat Apr 25, 2015 12:18 am (GMT 10)<br />
Topic Replies: 4<br /><br />
<span class="postbody">Hi,
<br />
You can make it simpler by ignore the pseudo code in ppt. Since the nearest neighbour will be sure to stored in the same leaf node which the query node is stored. Then you will find it's easy to get all the satisfied nodes from 3 trees then sort them against the distance. Hope it will help.
<br />
<br />
Cheers!
<br />
<br />
Sol
</span><br />
COMP9318 Assignments :: RE: Assignment 1 Question 4.4
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73582#73582
Fri, 24 Apr 2015 14:04:04 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=5087" target="_blank">SOLKI</a><br />
Posted: Sat Apr 25, 2015 12:04 am (GMT 10)<br />
Topic Replies: 3<br /><br />
<span class="postbody">Hi ldawes,
<br />
I also get confused.
<br />
Besides, is d = 20 the dimension of result matrix after AND-then-OR construction or not?
<br />
<br />
Cheers!
<br />
<br />
Sol
</span><br />
COMP9318 Assignments :: Assignment1 Q4
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73581#73581
Fri, 24 Apr 2015 13:59:28 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=5087" target="_blank">SOLKI</a><br />
Subject: Assignment1 Q4<br />
Posted: Fri Apr 24, 2015 11:59 pm (GMT 10)<br />
Topic Replies: 1<br /><br />
<span class="postbody">Hi,
<br />
Let me be quick.
<br />
In the Spec, "The dimensionality of the binary vectors are d = 20". Is this 20 the result of AND-then-OR construction or not? I'm confused with this for 1 day and if d is the dimensions of the original matrix, I mean the matrix before constructions, then the question 4.2 can not make any sense with r = 10 and b = 20. The product of r and b is 200, much larger than the dimension d which is only 20.
<br />
Hope someone could help, thank you!
<br />
<br />
Cheers!
<br />
<br />
Sol
</span><br />
COMP9318 Assignments :: RE: About FindNN Algorithm, i have a question
https://cgi.cse.unsw.edu.au:443/~forums/support/viewtopic.php?p=73580#73580
Fri, 24 Apr 2015 13:08:58 GMTAuthor: <a href="https://cgi.cse.unsw.edu.au:443/~forums/support/profile.php?mode=viewprofile&u=5084" target="_blank">wq1456728</a><br />
Posted: Fri Apr 24, 2015 11:08 pm (GMT 10)<br />
Topic Replies: 4<br /><br />
<span class="postbody">Thanks to all, i understand it ~
</span><br />