Monday, September 26, 2011

The Value Of Thinking About Rating Systems

A new project I have just started is going to generate personalized movie ratings for users. The way it works is as follows. You rate the movies you have seen. Then the system finds other users with similar tastes to extrapolate how much the you will like some other movies. It is currently written entirely in Python.
That's from Sergey Brin's 1996 resume. Prior, of course, to co-founding Google.

Correlation, causation, or aberration? You tell me. 

Tuesday, September 13, 2011

Can You Keep Your Job While Innovating?

Justin Lall says, at least in the context of competitive bridge, basically no. This is similar to NFL coaches not going for it on fourth down as often as they should, but a little bit more nuanced.
The situation in bridge is less dire, as percentage plays are respected, and the math is often easily demonstrable. However, a large part of becoming a great bridge player, good enough to be hired to play on professional teams, is having excellent judgement about when situations to go against conventional wisdom or A priori odds.... 
The biggest problem is that you are not just thinking about a bridge problem anymore, that’s the easy part, you’re deciding if making this play is worth possibly getting fired, get a bad reputation, maybe never get hired again and be forced to get a 9-5 job that pays much less.
You can generalize his argument to the claim that some now vs more later trade-offs are a superset of exploration vs exploitation trade-offs.

I agree. In the short run people who follow the status quo have an edge, especially in the absence of good long run indicators.

One prediction of this is that if 1) the football season were longer, or 2) instantaneous ratings for the probability that a team would win became standard (like "pawn units" in chess), more coaches would go for it on fourth down.


If you think this post was just an elaborate excuse to use the word "superset," you know too much.

Monday, September 5, 2011

Is One Willpower Trade-Off Fundamental?

 [T]he authors offer no systematic account of the trade-offs the brain must make among goals that differ in their likelihood of success, their time horizons and their evolutionary impact. The old joke about the man in front of a firing squad who refuses the customary last cigarette because he’s trying to quit reminds us that deferring a reward does not always make sense....
That's from Steven Pinker's otherwise mostly positive review of Baumeister and Tierney's Willpower. The relevant trade-offs in my current schema are some now vs more later (delay discounting) and sure bet vs shot in the dark (probability discounting).

Mentioning both in the same sentence reminds me of an old question of whether these trade-offs are actually equivalent on some fundamental level. Green and Myerson's highly cited '04 review addresses this.

From one perspective, you can think of a lower probability for success on a task as indicating that it will need to be repeated more times, and thus take longer. On the flip side, you can think of a longer delay as indicating a lower chance that the expected reward will actually be received.

One problem with both of these theories is that for humans, the discounting rate differs between delay and probability with respect to amount. That is, smaller rewards lose value more quickly as delay increases, but larger rewards lose value more quickly as the odds against increases. Check out this figure for some data.

This amount finding remains contentious, because it has not replicated in studies of non-human animals. And in those studies, it looks like delay discounting could be the more fundamental force of the two.

Still, some intuition remains on the side of considering them as distinct trade-offs. I don't think you'd want to tell an out-of-luck gambler in Vegas that if they just play longer, they'll eventually win.


After reading one of Baumeister's original papers, Tierney's exposition, Pinker's review, and Robert Kurzban's three published critiques, I am now in the curious situation of having devoured a fair number of words about a book without any inclination or intention to read the text itself.

Do I feel bad about this? Not really. I still think that the hegemony of the book is a bit anachronistic, and prefer the back-and-forth commentary that blog posts and edited articles provide.

But perhaps this is just a rationalization of my lack of willpower to actually close my computer screen and pick up a book.