ALL ART BURNS

I'll give a short presentation about the lasersaur and my build and a few of us will be around hacking on makerbots and whatnot. If you're interested in open source hardware, stop by and chat.

pyrotopia has a kickstarter page, help us raise funds to put on a free fire arts festival in Pittsburgh!

Let's say I have a symmetric correlation matrix where the row/column order is arbitrary. That is, for any given row or column, it doesn't matter what that row/column's neighbors are:

ABC
A
B
C

and

CAB
C
A
B

Are both legal and equivalent.

1) What is the name of this property where the order does not matter? At work we say "there is no locality", but I suspect we made that up.


2) What is the name of the operation where I sort the rows/columns by the strength of the correlation so that the row/column with the most or strongest correlations is on the left or top?

So:

A B C
A 0 .9 .5
B .9 0 0
C .5 0 0


becomes

B A C
B 0 .9 0
A .9 0 .5
C 0 .5 0

I'm about to be done with LJ. I'm ready to move all of my public posts to my blogs at allartburns.org or flatline.net and all of my "private" posts to facebook or google+.

The one issue being that I primarily use facebook for local to PGH events/people, which is why I'm not FB friends with my bro xiitone or my sis-in-law kittymonkey. It's not that I don't love all my non-Pittsburgh pals, it's that I really like FB being focused on local events, the way it was when I signed up "back in the day".

Suggestions/comments are appreciated, I'll probably kill this account within a month or so.

let's try this again.

Bosch PGH has an opening I'd like to know more about, if you know someone there, ping me? thx.

well, not as a real professor, but doing an el-wire workshop at HackPGH, our local hackerspace. Basically me lecturing or teaching soldering tricks for 4 hours straight. Tired, lost a few bucks on materials due to an under-subscribed class, but still happy I did it.

Fukushima is not Chernobyl, scientists repeat, and even Chernobyl was not as deadly as popularly believed.

Dire warnings of radiation spreading from Japan's embattled Fukushima Daiichi nuclear power plant to deadly effect across Japan, or even to California, are likely overblown, they say.

Radiation is all around us, varying with the number of miles we fly, the elevation of our towns, and the minerals in our environments, scientists point out. We live with it, and most of the time it is harmless.
[...]

More evidence that I didn't pay enough attention in stats class. Also, I'm not at liberty to say what the actual data is, which is why I'm making up silly examples about the content of these tables using food.

Also, I'm trying to find a good textbook on this subject but all I'm finding are generic stats books. What sort of keywords should I be searching for, or which authors/researchers should I be reading?



1) Someone gives me a correlation table of a population's favorite recipes that is sparse, non-diagonal, and has all the negatives filtered out. They claim it is non-diagonal because they filtered out positive correlations that made the matrix be non-sparse; and that it has no negatives because negatives are useless for our purposes (which is true, we don't care about negatives).

Q1A) Other than being non-diagonal, are there any other properties I should assume this table has or does not have? Does all the filtering make it no longer a "real" correlation table?

Q1B) Other than performance / memory limitations, why would I prefer a sparse table over a dense table? Why not keep all the weak correlations then filter by correlation level as I use the table?
(ex: give me only the top 10 things that go with cake or give me everything, no matter how weak, that goes with coffee.)

Q1C) How valid is this table? Why filter out a correlation between a popular food, say, bread, and half the table? If everyone who likes bread also likes half of the table, why would we remove that and why doesn't that invalidate the table?


2) I use a machine learning tool to build a correlation table between types of food based on their typical ingredients. For example, sheet cake would correlate strongly with cupcakes (mostly the same ingredients) but only slightly with a BLT sandwich (due to the flour in the bread). This matrix is diagonal and has only positive correlations.

Q2A) Do I want/need negative correlations? If so, how would I create them? Would I do something like normalize the table so that instead of being in the positive space it goes from say, -0.5 to 0.5? Or would I change my tool to generate negative correlations based on some distance function between the ingredient vector spaces?

Q2B) Similar to Q1C, is there a reason to go through and filter out weak correlations or to normalize the correlations to go from 0.0-1.0? I've got RAM and CPU to burn, is there a reason to not keep all the data?

A unionized public employee, a member of the Tea Party and a CEO are sitting at a table. In the middle of the table there is a plate with a dozen cookies on it. The CEO reaches across and takes 11 cookies, looks at the tea partier and says,"look out for that union guy, he wants a piece of your cookie".