00:08 < naywhayare> andrewmw94: friday is tomorrow, right?

00:08 < naywhayare> I can do that

00:21 < andrewmw94> ok, I'll see you then.

00:21 < andrewmw94> Bye.

14:07 < andrewmw94> naywhayare: Are you free?

14:10 < naywhayare> andrewmw94: yeah, just got in

14:11 < andrewmw94> ok. So I have a question about the desgin for the Hilbert-R tree

14:11 < naywhayare> okay, go a head

14:11 < naywhayare> *ahead

14:11 < andrewmw94> It's a bulk-loading method, so I assume it should be a constructor.

14:12 < naywhayare> right, I would agree with that

14:12 < andrewmw94> So I think I would add say a boolean variable to the constructor "BulkLoad"

14:13 < andrewmw94> but if we were to add more bulk loading methods later, that would be a bad idea

14:14 < andrewmw94> But I can't think of how you would do that. Possibly a function that provides an ordering of points would be used, and the constructor would be generalized to rely on that?

14:15 < andrewmw94> Do you think it's likely that more bulk-loading methods will be added later?

14:15 < naywhayare> my understanding of the hilbert r tree is very basic, so I may say something incorrect

14:15 < naywhayare> but do you think it would be reasonable to add another template parameter -- PackingPolicy -- that will take the tree and "pack" it into the more efficient ordering proposed by the hilbert R tree?

14:16 < naywhayare> for the standard R Tree, PackingPolicy is NoPacking, or, it does nothing

14:16 < naywhayare> but for the Hilbert R Tree, it would be HilbertPacking, where you map all your rectangles into this 4d space and then order them according to the hilbert curve (or something like that)

14:17 < naywhayare> or I guess they give like five ways to pack it... 4d-xy, 4d-cd, 2d-c, 2dz-c

14:17 < naywhayare> hm, that's four, not five

14:17 < andrewmw94> what do you mean by "take the tree and "pack" it into the more efficient ordering"?

14:18 < naywhayare> I thought that's what the hilbert R tree was

14:18 < naywhayare> let me do a bit more reading

14:18 < andrewmw94> My understanding is that the tree is created by this, not that it is modified

14:18 < andrewmw94> Or, I suppose it could be modified, but I think it would be easier to implement as the tree being "created"

14:21 < andrewmw94> So, supose you have a set of 100 points. You order them according to the Hilbert curve. Then let's say you have four children per node. You break that into four groups (0-24, 25-49, 50-74-75-99) and put each group into a different child node.

14:21 < naywhayare> okay, so what I misunderstood was that "data rectangle" did not refer to an already existing tree node

14:22 < naywhayare> the construction algorithm is Figure 7, right? (from "Hilbert R Tree: An Improved R-Tree Using Fractals")

14:22 < andrewmw94> Yeah, I think we decided that we're only going to worry about points until we can think of a ML application for hyper rectangles

14:23 < naywhayare> hm, okay, here is a modified thought which is less wrong, then --

14:24 < naywhayare> add an additional template parameter, but call it "BatchConstructionPolicy" or something like that; basically, it's what gets done to insert points when many points are given (such as in the first constructor)

14:24 < naywhayare> for a regular R tree, this is just some wrapper that sequentially calls Insert() (I think that's what happens, anyway)

14:25 < naywhayare> and for the Hilbert R Tree we can have some HilbertBatchConstruction class that takes an input matrix and implements that batch construction algorithm

14:26 < andrewmw94> ok. I think that will work

14:26 < andrewmw94> More templates. Hurrah :)

14:26 < naywhayare> yeah, I think that might be reasonable

14:27 < naywhayare> I haven't read the section on the dynamic Hilbert R Tree, but I think whatever's presented there could probably fit into the existing SplitType/DescentType abstractions

14:27 < andrewmw94> I think so. They're pretty general.

14:28 < naywhayare> also, some time back I found a bug in the X tree code that I haven't completely tracked down yet: http://www.mlpack.org/trac/ticket/368

14:28 < naywhayare> I think I CC'ed you on the ticket, so assuming Trac is working correctly, I guess you should already know about it

14:29 < andrewmw94> yeah. I think there were two bugs actually

14:29 < naywhayare> anyway, I haven't had a chance to look further into it, and it may be a while until I do, so if you find it interesting, feel free

14:29 < naywhayare> yeah, another one is a visual studio bug

14:29 < andrewmw94> Yeah, that's what I thought I remembered. I don't have VS, so that could be harder for me to fix.

14:29 < naywhayare> I think there's a workaround in place for that one for now, so it's less important

14:30 < andrewmw94> I suppose I can get it through DreamSpark or whatever it's called, but I hate programming on Windows

14:30 < andrewmw94> Yeah. I'll look into the first bug sometime soon. I've been busy with starting a chess club here and things like that

14:31 < andrewmw94> the amount of paperwork required was ridiculous

14:31 < naywhayare> yeah; for a long time I ran the Linux Users Group here at Tech and it was a large amount of paperwork

14:32 < naywhayare> had to renew chartering with the student government association every year, had to have bylaws updated every year, signatures from all kinds of random people

14:32 < naywhayare> I'm very glad I don't do that anymore because it's only become even more busy work

14:32 < andrewmw94> Yeah, I should probably find some naive person to do it next year...

14:33 < andrewmw94> I'm getting kind of nervous with how accurate Dilbert is. I'm worried what a real job would end up like...

14:34 < naywhayare> I did some work for a fairly standard Dilbert-esque company over the past year... yeah, it was quite accurate

14:35 < andrewmw94> I assumed it was completely fabricated, then I came to college and it seems more and more accurate as I get more and more experience. GSoC was an exception though :)

14:35 < naywhayare> http://dilbert.com/strips/comic/2006-03-18/ always entertained me

14:35 < naywhayare> applicable to my life too :)

14:36 < naywhayare> I did an internship at Google in 2010, and it was very much not the Dilbert experience

14:36 < andrewmw94> I really like this one: http://dilbert.com/fast/2010-05-01/

14:36 < naywhayare> so there are definitely exceptions, but man, I've also been some places that suffered from PHB syndrome

14:36 < naywhayare> haha

14:38 < naywhayare> haha, wow, this is a good one: http://dilbert.com/fast/2010-05-05/

14:38 < naywhayare> that was basically every day at the company I worked for last year ("how accurate do you think the model will be?" "how long do you think it will be until you can make the model reach X accuracy?")

14:39 < andrewmw94> http://www.dilbert.com/fast/2011-05-26/

14:39 < andrewmw94> I spent almost an hour trying to find that one

14:40 < andrewmw94> But then I found this website using google rather than the dilbert search: http://www.bfmartin.ca/finder/strips/2011-05.html

14:40 < naywhayare> haha, yes! that was the entirety of my undergraduate experience with group projects

14:40 < andrewmw94> I guess the commic is incorrectly tagged as dilbert rather than dogbert. And I kept using dogbert in my searches.

14:41 < andrewmw94> Yeah, working with a group is terrible. I'm still not sure what it was supposed to teach me

14:41 < andrewmw94> perhaps that most people are stupid?

14:42 < naywhayare> at least in my case there were often big skill and experience inequalities

14:42 < andrewmw94> Yeah. It's always easier to do everything myself than it is to explain what needs to be done to the other students

14:43 < naywhayare> once or twice I managed to work in a group where we were all on equal footing and it was great

14:43 < naywhayare> but that was a rare experience

14:43 < andrewmw94> well, not always. I made a chess engine with another student who is good.

14:43 < naywhayare> ooh, a full-on chess AI, or a rendering engine of some sort?

14:46 < andrewmw94> a full-on AI. It's not good, but it does play chess.

14:46 < andrewmw94> The rendering engine should be this semester. I'm not sure whether I'll try to make a ray-tracer or something else yet.

14:48 < andrewmw94> But the ray-tracer is really interesting. My family has background in physics (Mom and Dad are both Ph.D. physicists and my Grandfather is a physicist who speciallized in optics)

14:49 < naywhayare> neat; I've been meaning to learn more about ray tracing recently

14:49 < naywhayare> kd-trees are often used in ray tracing, from what I understand, to quickly find the interactions between rays and objects in the scene

14:49 < andrewmw94> Yeah, I like how it's elegant in the mathematical sense.

14:49 < andrewmw94> yes. We went over them in class the other day.

14:50 < naywhayare> I can't say I'm more familiar than that; I was mistakenly an EE undergrad when I should have been CS, so there are lots of standard CS ideas that I don't know very well

14:50 < andrewmw94> They can be used for several things, dividing up the scene to make rendering faster, keeping track of which objects block other objects, etc

14:50 < andrewmw94> ahh, I had a bit of trouble with EE.

14:50 < andrewmw94> VHDL was not my cup of tea.

14:50 < naywhayare> VHDL is a terrible language; I don't think it's anyone's cup of tea

14:51 < andrewmw94> I wrote a tic-tac-toe AI for it (actually I wrote a C program that wrote the AI)

14:51 < naywhayare> verilog is better (and more commonly used in non-government industry), but not much in my opinion

14:51 < andrewmw94> but the compiler or the circuit was broken. You could verify that the lookup table had the correct move and possition, but the circuit wouldn't play it.

14:51 < andrewmw94> possibly a timing issue.

14:51 < naywhayare> I found the generics support in VHDL to be particularly bad, but it's been too many years so I can't completely remember the exact details of what I disliked about it

14:52 < andrewmw94> yeah. We also had a professor who thought he needed to teach us binary but assumed we knew VHDL.

14:52 < andrewmw94> It was painful.

14:53 < andrewmw94> But I have class in about 15 minutes and I still haven't finished my translation, so I should probably go. Thanks for the help.

14:53 < naywhayare> ah, see you later

14:53 < naywhayare> good luck! :)