« October 2006 | Main | December 2006 »
November 30, 2006
I Taught Him His Crossover, Too
Apparently Tony Parker heard that I made Ruth a formal proposal yesterday and said, "Oh shit, it looks like the cool kids are all getting engaged today." Otherwise, why not do it on an off day?
Speaking of our engagement, I should write down how that happened. For the kids, and all.
The thing is, if you've been paying attention, you know that the wedding is already halfway planned. If you're in the loop, you know that Ruth even had a hand in picking out the engagement ring.
Given all of that, you might have been surprised that I even made a "formal proposal." If so, you clearly don't know enough girls. Whatever else has already happened, no girl wants to be robbed of her one good chance to have someone on bended knee with a ring.
The problem was, it seemed to me like this would only be fun if it was a surprise, and after we got the ring from her grandmother over Thanksgiving she had to be expecting it at any minute. Any kind of proposal during a traditional romantic moment -- our anniversary dinner is around the corner, and there are tons opportunities for scenic walks on the beach -- was right out. Way too obvious. I thought about it a lot, and I could only come up with one time where I felt like it wouldn't be seen coming from a mile away: during a class.
So, I got in touch with one of her professors and arranged to add an extra slide to the end of the PowerPoint slide for her lecture on Wednesday. Then I waited outside of the class for forty-five minutes until the class was over and she put my proposal up:
Then I went in and did the knee thing and put the ring on her. All of her good girl friends are in the class, which was a nice bonus. They all seemed to get a kick out of it.
Meanwhile, here is the ring. If it were made of kryptonite, it would be able to take out Superman at fifty paces.
Posted by todd at 8:41 PM | Comments (7)
.9446
Bagging was surprisingly helpful. Our results now beat the original Netflix algorithm on the qualifying set, but just barely miss the cutoff to qualify for a progress prize. Which doesn't really matter, because we're lightyears behind the leaders.
The moral of the story is that the original Netflix algorithm can't be that great. It's really hard to believe that it was ten years in the making.
Posted by todd at 4:26 PM | Comments (2)
November 29, 2006
.9612
Pretty "eh" results from our first official submission. Tomorrow we try bagging some mediocre results and hope that it turns out miraculous. Don't hold your breath.
Posted by todd at 11:12 PM | Comments (0)
November 28, 2006
Encouraging Results
Turns out, each of us had a different, critical bug in his error calculations. We actually aren't doing that bad.
Posted by todd at 6:02 PM | Comments (1)
November 27, 2006
Disappointing Results
It looked for a while there like we might have an algorithm which produced competative results on the Netflix problem. However, it turns out that we in fact have a method which learns to do worse and worse until it looks like a random solution. How we managed to accomplish this is a total mystery.
Posted by todd at 11:25 PM | Comments (1)
November 26, 2006
That was Fun
It's really great to enjoy the company of your mate's family, isn't it?
Posted by todd at 10:30 PM | Comments (0)
November 25, 2006
A Note to the Neighbours
Dear Tina D. B.,
Thanks for letting us use your wireless all weekend. Without you and Wikipedia, at least three arguments would have gone on much longer than necessary.
Posted by todd at 5:32 PM | Comments (5)
November 24, 2006
The Best Thing Ever
Tuna steak, medium rare. Hands down.
Posted by todd at 4:51 PM | Comments (0)
November 23, 2006
I'm in Your Interwebs, Killing Your Memes
Has everyone already seen thefunniest.info?
Posted by todd at 7:07 PM | Comments (1)
November 22, 2006
Back on Sunday
We're in
Posted by todd at 8:57 AM | Comments (2)
November 21, 2006
Note for French People
From now on, it's cool to kiss my girlfriend on each cheek once per party. Any more than that and you will be understood to have invited da ruckus.
Posted by todd at 11:01 PM | Comments (0)
November 20, 2006
Sick!
I'm feeling a tad under the weather. Fortunately, the weather's nice, so it's not too bad. Unfortunately, the final for AI is tomorrow, and no one wants to sniffle through an exam.
Posted by todd at 11:39 PM | Comments (0)
November 19, 2006
This One's for Bo
California is awesome. One day last week I started to miss Massachusetts -- I was feeling nostalgic for the historic creepiness of the Northeast and the comfortable snowiness of our trip to Salem for Halloween last year. I went so far as to believe for a few hours that walking to and from the train to work in the snow every day all winter had been fun, rather than incredibly frustrating.
Later that evening, I went for a run outside, and only as I was turning back onto our block did I realize that this was the first thing that started to upset me about winter every year -- by now, at Bard, or in Waltham, I would probably not have felt much like running outside after dark. It's easy to take the weather here for granted, and to think you miss something less monotonous, but the one tone is fucking sweet.
For instance, yesterday -- that's November 18, if you're keeping score at home -- I spent the morning at an outdoor farmer's market, the afternoon playing basketball outside, and the evening on the beach grilling salmon and roasting marshmallows for smores with a bunch of neuroscientists and my machine learning homies. It was outstanding, and I'd be hard pressed to trade it even for a couple of interesting coffeehouses and a good used book store.
Posted by todd at 10:35 PM | Comments (11)
November 18, 2006
Fuck You, C++
At the moment, lines 70-73 of my Netflix code look like this:
while(true){
numcls = (rand()%40)+10;
cout << "Number of clusters: " << numcls << endl;
kmeans(numcls, usercount, (maxmov-minmov)+1);
}
If I compile and run my code, I get a segmentation fault in a method called at line 60. However, if I comment out the while loop at line 70 -- so that K-Means runs, but only once -- everything works fine and a I actually get decent results.
This would never happen in a real programming language.
(By 'this' I mean both the inexplicable crashing and the good results -- the code would be too slow to get any results.)
Posted by todd at 10:16 AM | Comments (1)
November 17, 2006
Oh, Snap!
Er, mulligan?
Posted by todd at 6:54 AM | Comments (7)
November 15, 2006
John Hollinger: Master Statistician
ESPN's resident statistician wants to see if he can make any interesting claims about the affect of the new ball on the game. Here's what he comes up with:
The most interesting data, however, come when the ball gets a little closer to the rim. Let's start from the foul line, where the league has improved from 73.3 percent to 74.5 percent. This is an impressive jump considering the league leader in free-throw attempts, LeBron James, has been notably inaccurate from the stripe.[...]
This is the most compelling evidence to date that the ball does tend to offer a kinder bounce to shooters. That said, let's be realistic about the scope of the difference: We're talking about one free throw out of 100.
So, you're calling that 1.2 percent change "an impressive jump," and "compelling," while also pointing out that it's not significant? What are we supposed to think? Also, what is the variance in "free throw percentage in November" like, year-to-year? I'm guessing it's way more than 1%, which means that you can't make any interesting claims at all about the difference between this year and last, much less any claims about the relationship between the new ball and scoring totals.
(In case you're wondering: yes, I am just bitter that I can't get Insider content.)
Posted by todd at 11:23 PM | Comments (1)
November 14, 2006
Since I'm Sure You're Waiting with Bated Breath
I did OK. Not great, and I'm somewhat disappointed with the kind of error I made, but I did OK.
Also, UCI basketball home games are a hoot.
Posted by todd at 11:45 PM | Comments (0)
November 13, 2006
Apprehension
First exam-like experience of graduate school is tomorrow, so I'm a bit too distracted to post. Meanwhile, if you can think of a good mneumonic device for the derivation of the stochastic gradient descent rule for backpropagation of error in neural networks, please leave a comment.
Posted by todd at 11:22 PM | Comments (2)
November 12, 2006
When You Register [she]and[he].[com|net|org|gov], It's Official
So, you know, keep an eye on Ruth & Todd (dot) net for ... important updates? Personal histories? I really have no idea. Anything could happen at this point.
Posted by todd at 11:44 PM | Comments (5)
November 11, 2006
How the Locals are Like Children and Todd is Like Bill Cosby
It was 10pm , and we were standing in the lobby of a theater, having just finished a long movie (Babel, which was enjoyed by all). People were hungry, and in need of a place to sit and talk.
"Too bad this is Southern California," I said, "And there are no diners for hundreds of miles. This is diner-time."
"Oh, we have diners," said a life-long resident.
"Really? I haven't seen any."
"Yeah, we have Denny's, and ..."
Unfortunately, I'm not sure what she was going to say next, because I laughed in her face. This has happened several times since we moved here, and I feel bad every time, but Southern Californians say the darnedest things.
Posted by todd at 10:05 PM | Comments (5)
November 10, 2006
Looks Like All Three are Leaving with One Guy
Damnable Feynman strikes again.
Posted by todd at 11:35 PM | Comments (1)
November 9, 2006
Mmmrrrow!
Here's an article about how cats survive 6 to 40 story falls.
Got that from Bo, but I'm too lazy to look up the permalink right now.
Posted by todd at 11:40 PM | Comments (0)
November 8, 2006
At Least He Didn't Get Tenure
News flash for Adrianne and Tony! The Onion reports that Area Man Achieves Your Dream. (Via Gaiman.)
Posted by todd at 9:11 AM | Comments (1)
November 7, 2006
Don't Tip the Vote Over, Baby
Stood in line for a solid hour today, but damnit I got my sticker.
I miss the voting place in Barrytown, though. The last presidential election was one of the best days of senior year, with its delicious chowder and ridiculous apple pie and the mile long walk home through fall foliage.
That is, until the results came in and we drank until I puked. Haven't looked at vodka the same way since.
Posted by todd at 10:30 PM | Comments (0)
November 6, 2006
More Geeky Public Service
The .emacs preference file which ships with the Ubuntu package for Gnu Emacs really sucks. It defaults to all of the settings which no one would ever want: no syntax highlighting, no autofill in text modes, small windows, and scrolling with a mouse wheel doesn't work. The editor constantly asks if you'd like a newline to end your file, and there's no visual feedback for selected regions. It's just completely ridiculous.
Here is what I consider a tolerable bare-bones configuration, posted so that I never have to look up those functions again, and in case it helps anyone else.
Posted by todd at 10:54 PM | Comments (0)
November 5, 2006
Saddam to be Asphyxiated
I see (via PZ ) that Saddam Hussein has been sentenced to death by hanging in an Iraqi court.
Death by hanging? People still do that? That is some fucked up shit. I know the guy is a bad man, but don't you want to move away from a culture where the government sentences people to gruesome deaths?
Posted by todd at 10:29 PM | Comments (2)
November 4, 2006
Well, That was Awesome
First time since the sixth grade that I attended an NBA game in which the home team won. Good times.
Posted by todd at 11:14 PM | Comments (0)
November 3, 2006
Come to Think of It, I Don't Know
ABC reports that the government apparently wants singles to wait until they're at least thirty to have sex, which prompts a very good question from one of the website's sponsors:
Seriously, thirty? What are you waiting for?
And that may be the first time I've ever thought a Viagra ad made sense.
Posted by todd at 5:10 PM | Comments (0)
November 2, 2006
Who Took the Bomb?
Were you aware that there is a video for "Deceptacon," and that it features an awesome choreographed dance? You are now!
(If my endorsement isn't enough, know that I showed it to the DRZA and she called it "teh happy." High praise!)
Posted by todd at 11:32 PM | Comments (3)
November 1, 2006
Awesome
First I posted some code as an excuse for having no post written, then I discovered a bug in the code, so now I have nothing at all. But check it out: factor graphs are awesome. They're more expressive than Bayes Nets and Markov Random Fields combined! Oooooh.
Posted by todd at 11:20 PM | Comments (5)