better know an insect

I've decided I should know more about bugs. Partly cuz my job these days is to watch flies. That is what a PhD qualifies you for, you know. So, here is the first installation in my 4 million part series, better know an insect, in which I sit down with one Wikipedia article about some kinda bug. Today's subject: the roly poly, aka the pillbug, aka woodlice of the family Armadillidiidae.

We all know, of course, that roly polys can roll up into little balls to protect their tasty insides from predators. Here is a surprising factoid: roly polys are not insects! They're crustaceans. Count them legs! 14! Wow, who knew? Not you! Don't you pretend you knew that!

What else did I learn? From the Beeb: "Woodlice eat rotting plants, fungi and their own faeces, but they don’t pee!" Awesome! Oh, and they can live for 3-4 years. That is a long friggin' time! Man, I should get me some pet roly poly's.

Above painting by Jean Bradbury taken w/o permission from here. Cool painting, Jean!

cap'n bringdown rocks out to the classics

Here are some photos I took about an hour ago of Cap'n Bringdown rocking out to Beethoven Symphony 9, conducted by Toscanini, a YouTube video I found linked from an NPR show entitled ... something about a treasure trove of classical something or other on YouTube. We were trying to become cultured. We were previously familiar with Toscanini from his delicious delicious sandwiches ... mmm toscaninis.

my evil little baby

Last year, I caught a praying mantis on the door to my apartment. We kept it in a cage in our lab and fed it live insects that we would hunt for around the lab and campus. Mostly, she got crickets and some kind of shield bug that seems fairly common in the area. Once, she got a giant grasshopper. That was pretty gross. She was an awesome pet cuz she just looked so evil when hunting her prey. She liked to hang from the top of the cage. I think she could only see motion, not still insects. But when the little bugger would move, she would stay absolutely still, except for her head, which would turn all Exorcist-like to stare. Then, she would creep very slowly until she was just above her prey, then pounce and grab it in her claws. Even though she was only about 4 inches long, I was kinda afraid of her. She would eat her food live, while it struggled in her claws. She would start at whatever part was most convenient, as the bug's struggle was pretty futile to her. She would eat every part -- antennae, legs, brains, ooey-gooey guts -- except for whatever excrement remained at the core of the bug. Here is a fact I'm sure you wanted to know -- the half-eaten shield bugs had a grassy, herby, sweet smell, not at all unpleasant; this is strange cuz shield bug is synonymous with stink bug. Maybe I'm weird, I didn't think they were all that stinky.

My mantis died last fall a few days after laying a really nasty looking egg sac. Apparently, the eggs hatched this spring and a couple manti survived briefly, but we were not around that weekend to feed them and they all died. The picture above shows her eating one of the above discussed shield bugs. Isn't she precious? This is the first meal we fed her. We were ecstatic that she had caught the bug we threw in her cage. RIP, my evil little baby.

jebus cripes supertard

I am maybe making a t-shirt out of this picture. Maybe. It may be too random. but I thought it was pretty. It is mostly stolen from
The Searcher's Flickr stream.

In the original, it is supposed to be a coloring book picture. Check out the original, it has funny tips on what colored crayons one should use. Also, there's other good stuff there. Anyways, the literalists out there have to mesh dinosaur fossils with the earth being a few thousand years old, hence the idea that Jesus may have coexisted with dinosaurs, and the term "jesus horse". Not trying to be mean-spirited, I just like the picture. Jesus & his dinosaur look so poised for a heroic deed.

Presents!

Presents are fun, cuz everybody likes surprises. Not big, life-altering surprises, but tiny little surprises that are just a tiny spark of excitement in one's otherwise dull existence. Anyways, if you, like me, would like a tiny spark of excitement on more than two days a year (your birthday and Jesus' birthday), do it yourself! Apparently, at the "Something Store" you send them $10 and they send you a surprise ("something", to use their term). Anyways, I will be saving this special little something for some extra-dark time in the future when I will *not* be changing addresses within the next two weeks (I'm moving one block west at the end of the month :)).

orchid calligraphy

I got a tablet for my computer, and I've been playing with it some this week. Here is what I got so far:

Portishead Rip Video

Portishead - The Rip

Beautiful song with beautifully matched video. Really captures the feel of the song.

Werewolf Bar Mitzvah

Here is a link to the entire song "Werewolf Bar Mitzvah", a short clip of which was in an episode of 30 Rock.

Dictator Style: Lifestyles of the World's Most Colorful Despots

Dictator Style: Lifestyles of the World's Most Colorful Despots
by Peter York

I'm not going to buy or in any way go out of my way to find this book, but I would definitely leaf through it if I happen to see it lying around! (Quite the endorsement?) Apparently, it is a picture book showing the art and architectural tastes of recent history's worst dictators. It's certainly a catchy gimmick. Anyways, shock and awe, some of these dictators had rather poor, grandiose tastes. Read more about it at Head Butler or Bibliostyle or Amazon.

Here are some pictures from Amazon:

Painting hanging in one of Saddam Hussein's palaces.

Jean Bedel Bokassa's cool eagle throne.

drawrings

Who is More Liberal, Obama or Clinton?

It seems that the policies of an individual congressman are well-described by a single number, the first dimension of the two-dimensional DW-NOMINATE score. Yay America!

In the study Who is More Liberal, Obama or Clinton?, the authors histogram the DW-NOMINATE score for all members of congress. Their point is to show that Clinton and Obama are statistically identical. Of course, this isn't terribly surprising, since there is a way senators must vote in order to be electable. Funny that I still feel pretty strongly about who I want to win.

It's also interesting to see how well-separated the Democrats and Republicans are. I think the animation (above) showing the parties clustering over the past few years is entertaining.

Cross Entropy Method

I'm always forgetting the name of this algorithm, and it suddenly came to me this morning, so I thought I'd write it down here so that I can remember it. Ooh, it's on Wikipedia. There's a more complete tutorial here.
that maximizes a given score function which has many local optima. While algorithms like simulated annealing keep track of the current estimate of , the CE method keeps track of a distribution over parameterized by a vector , . In practice, we initialize so that the distribution has a high variance, and the algorithm decreases this variance. Usually, our distribution is something like a Gaussian with diagonal covariance.

The algorithm requires an initial vector of parameters describing the initial estimate of the distribution of . Usually, we will set this distribution to be centered at our initial best guess of and to have high variance. Here is pseudocode for iteration of the optimization:

1. Generate sample parameters:

2. Score the samples:

3. Choose the with the $M$ highest scores, call these
.
4. Choose the maximum likelihood parameter for generating this set

5. Store in the score for the worst elite sample,

6. If is not better than , or the set has low variance, we have converged.

Here is a Matlab function that does this: cross_entropy_method.m.

Suppose that we were not setting at each iteration to a different value, but instead were defining elite samples as those above a fixed . Then this algorithm can be viewed as finding the parameters that minimize the KL divergence between the distribution

and . The above algorithm can be seen as finding the sequence of both and that converges to the optimal value.