"Avoid temporary variables"??? (A lot of the rest is reasonable, but could be subsumed into a reasonable version of 'lint' for whatever language you're working in. Is there an R lint?)

Read from top-to-bottom.

"The din of younger colleagues tapping keyboards is never soothing, but sitting in the press room of the Ideas Forum felt like a human rights violation. What could anyone write about something so tyrannically dull— other than an angry elegy for the massacre of meaning?" --- A little purple, but still pretty funny.

so, two comments:
(1) "computable web" != "canonical names." Common mistake.
(2) the "ambiguity" of reference that [some of] the semantic web people are working to eliminate here isn't the ambiguity he's describing ("My ... apartment has been described as being in the Lower Haight, Duboce Triangle, or Upper Castro, depending on who you ask..."), where one "thing" can have multiple names -- but the exact opposite, where one name refers to many (different) things, in different contexts. Imagine if "Duboce Triangle" was the name of a neighborhood *and* a newspaper about that neighborhood, *and* also the collective name for all the people living within 2 miles of Warden's apartment. A person might even use the same noun (or noun phrase) to refer to all three things, within the space of a single unit of text. It'd get pretty confusing. Using "canonical" names is a (admittedly somewhat simplistic) attempt to get around *that* problem, rather than the one he's describing; and saying that you "embrace" the ambiguity that's latent here is equivalent to saying that you don't care if the web is unusable to certain groups of people (e.g. scientists, researchers) who *are* concerned with avoiding this sort of ambiguity. "People searching for movie times" is just a test-case for "people searching for data about a 'gene'."

Also, I love someone who writes critically about Wolfram (.data, Alpha, and all the rest) as much as the next guy -- but saying, "the web is written for humans to read" is pretty laughable when it's not coming out of the mouth of a guy named "Firefox." The web is written for your web browser, and no number of SXSW presentations will change that.

</rant>

"We now propose a method for learning the mixture components given n i.i.d. samples y_n
drawn from a graphical mixture model P(y). Our method proceeds in two stages. First, we estimate the graph G_∪ := U_{r}^{h=1} G_h, which is the union of the Markov graphs of the mixture. This is accomplished via a series of rank tests. Note that in the special case when G_h ≡ G_∪, this also gives the graph estimates of the component models. We then use the graph estimate hat{G}_∪ to obtain the pairwise marginals of the respective mixture components via a spectral decomposition method. Finally, we use the Chow-Liu algorithm to obtain tree approximations {T_h}_h of the individual mixture components." -- To do: review how this works in the context of gene expression experiments for transcription factor regulatory relationships, which are (presumably) mixtures of a couple different underlying models or modes.

Where I come from, we usually call this "yak shaving."

Zellig Harris! Sweet!

Ah, this is the "Join Bayes Nets" paper, reworked...

"Stars are WRONG" -- (so it would be a bad idea, to print a few of these onto cards and start handing them out at train stations?)

"Charles Babbage described a computer powered by a steam engine; we de- scribe a heat engine powered by programs! We admit that the significance of this line of thinking remains a bit mysterious."

The list of things that I need to blog about grows ... longer.  

"If you skip this step—even with a downloaded library—you’re a baaaaaad genetic programmer. Turn in your Jaws and go back to machine learning land."

Norbert Weiner is ROFLing in his grave.

"Using “individual histories”—temporal sequences of all reproduction events and phenotypic changes of individuals and their ancestors—we present an alternative approach to quantifying selection in diverse experimental settings..."

"We consider the problem of estimating the topology of spatial interactions in a discrete state, discrete time spatio-temporal graphical model where the interactions affect the temporal evolution of each agent in a network."

"Through an extensive simulation study, we show that a simple modification of a method relying on a Gaussian approximation achieves good performance and is very fast. "

"We prove identifiability of parameters for a broad class of random graph mixture models. These models are characterized by a partition of the set of graph nodes into latent (unobservable) groups. The connectivities between nodes are independent random variables when conditioned on the groups of the nodes being connected. In the binary random graph case, in which edges are either present or absent, these models are known as stochastic blockmodels and have been widely used in the social sciences and, more recently, in biology. " -- To read, in the context of the blockmodeling paper from a few weeks back.

To think about, in the context of RDF and query.

That's Jun Liu, and therefore, worth reading.

Don't I already own this book...?

Really excited to read this!

Fascinating ... and, I admit, a little creepy. I say this as someone who accidentally played four hours of video games last night. (Suddenly I looked up, it was 1am, and I was all, "whoa...?") John, this goes back to the "washing the dishes" game we were talking about a long time ago? Also, I kinda wonder: sure, it's creepy when a *game* does this to hook you. But I kinda wonder if some of these psychology-based game-honed ideas couldn't be put to use in a classroom setting with students. How could that work? Tokens? Medals? "Coins? (a la Mario?) Lots of little questions? Probably effective teachers, especially at the grade school level, already use a combination of these techniques...?

"I argue that a priori symmetries need never constrain our probability attributions, even for initial credences."

"I certainly don't mean to suggest that the ancient Picts generated their petroglyphs using throws of 7d6." -- Liberman reveals his background as a D&D-playing nerd (one of us, one of us....)

"You can come out when you can properly explain the differences between Modernist architecture and postmodern ornamentation."

Probably should have bought and read one of his books (Scientific Explanation and the Causal Structure of the World) a long time ago.

"Here, we propose a model of individual e-mail communication that is sufficiently rich to capture meaningful variability across individuals, while remaining simple enough to be interpretable. We show that the model, a cascading non-homogeneous Poisson process, can be formulated as a double-chain hidden Markov model..."

"In other words, the more prestigious the journal (as measured by its "impact factor"), the less likely the genetic association studies it publishes are to be replicated." -- a great line. But to be honest, I'm less concerned by this kind of thing today than I was (probably) a year ago. First of all, I expect it from Wade. But second, I don't have that high an opinion of the vetting process that goes into journal articles. Peer review is not the arbiter of truth, it's the beginning of the investigation -- plenty of carefully-worded refereed journal articles (*especially* in Nature & Science) are also over-ambitious, breathlessly phrased, and written to attract (more) funding. Whatever the business of newspapers is, it should be clear that the business of [some] journals is not science, but *surprising* science. That's often a different thing.

I've said it once, I'll say it again -- *You Cannot Trust the NYT Science Writers.* They're not in the business of explaining science; they are in the business of sensationalizing science for the readers who happened to have wandered off of the NYT Fashion section and are wondering where they are.

"R for ... SPSS users." Need to send this to Rachel...

A reasonable tutorial? I still think that some of the local-linear-embedding techniques would work well in the context of disk or storage layout in a database setting.

These are the Bell Numbers (http://mathworld.wolfram.com/BellNumber.html), no? Or are the letters ordered in some way (alphabetically, or something)?

What's the old adage about wrestling with a pig? At any rate, a reasonable takedown of Kevin Kelly: "If there is something new in the writings of Kelly and other cyberprophets, it is the fact that they don't just predict a future in which the curves slope the wrong way, they endorse it. That is, along with the gee-whiz pronouncements about how the economy supposedly works goes a pronounced libertarian bent, a belief that the new economy is too dynamic, organic, or whatever to be regulated from above."

"Amateurs bet only once but professionals bet frequently."

Before I forget...

"We show that our arbitrage measure is invariant under changes of num\'{e}raire and equivalent probability. Moreover, such measure has a geometrical interpretation as a gauge connection." -- Ahhh.

To buy. (Descriptions of identifiability, and degrees-of-identifiability, in statistics).

I want to look at this a little more closely after the thesis defense, but my initial thoughts are that this doesn't seem to control very well for the exact *date* of publication, right? (Besides a quick note in the paper that most duplicate papers appear to be published in the same year, a quick scan reveals no other mention of publication time -- but maybe I missed something.) I wonder how often the publication in the "higher impact factor" journal is actually the *first* publication, chronologically.

For some reason, I keep seeing Nir Friedman sitting outside of the Starbucks down the street from me, talking to people I know. It was E.F.'s former post-doc, last week.

Can I ask a question? Is it the case that Brad DeLong's "Rosencrantz and Guildenstern Flip a Coin" example (http://delong.typepad.com/sdj/2009/03/cosma-shalizi-takes-me-to-probability-school-or-is-it-philosophy-school.html) is a particular instance of the example given in the last paragraph of this paper? (the answer, I am subsequently told, is "yes.")

The rest of the criticism may be (probably is) mostly on-target, but his criticism of McArdle's potted one-graf description of academic research vs. pharmaceutical work misses the mark pretty widely. Far from being "laughable," I'd say it's actually a pretty reasonable 100,000 ft. analogy, and I'd be truly surprised if Sue Lindquist or anyone else in her lab disagreed with it. What are laughable are the little blurbs made by his (Levenson's) students at the link -- for instance, "Scientists can cure Parkinson’s Disease in yeast – can they extend this to humans?" To the extent that the Lindquist lab will "cure Parkinson's" in humans, it will be in a theoretical sense. They won't be producing any drugs in our lifetimes. Also, "researchers pounding molecules into receptors" is a pretty poor description of the WIBR, and yeah, I can throw a stone from my office and hit them too. Grrrrr.

To be bought, read, and stored next to the Uri Alon books on my bookshelf.

Get out of my head, Jorge Cham!!

"When discussing complex systems, like brains and other societies, it is easy to oversimplify: I call this Occam's lobotomy." -- Awesome.

"It is clear that the choice of the initial state A is critical to this calculation. It is just not true that if the EV of a change from A to B is negative, then the EV of a change from B to A is positive. The choice of the initial state A is critical and always made very casually when attempting to calculate EVs."

"In this paper, we lay out the problem of discovering_functional communities_, and describe an approach to doing so. This method combines recent work on measuring information sharing across stochastic networks with an existing and successful community-discovery algorithm for weighted networks. We illustrate it with an application to a large biophysical model of the transition from beta to gamma rhythms in the hippocampus."

"One [way of studying market structure] looks to "industry characteristics" to explain why different industries develop in different ways; the other looks to the pattern of firm growth within a "typical" industry to describe the evolution of the size distribution of firms. In his new book, John Sutton sets out a unified theory that encompasses both approaches, while generating a series of novel predictions as to how markets evolve."

"The conclusion, boys and girls, should be that trade - voluntary exchange - is the exception rather than the rule and that markets are inherently and hopelessly incomplete. Live with it and start from that fact. The benchmark is no trade - pre-Friday Robinson Crusoe autarky. For every good, service or financial instrument that plays a role in your ‘model of the world’, you should explain why a market for it exists - why it is traded at all. Perhaps we shall get somewhere this time." --- Buiter's criticism of (among others) Robert Lucas.

"... empirics, as opposed to speculative theory and stylized facts, of economic growth..."

"In a review article about agent-based models Dietrich Stauffer once wrote "Physicists not only know everything, they also know everything better."" -- They're more like a contagion (physicists, that is) than anything else. Having done physics as a physicist would, they then spread out to new, uncharted (to them) areas, to explain how science should be done. Economics? It's really physics. Biology? Start with theoretical models, as a physicist would. Bioinformatics? We've solved those problems already. Philosophy of Science? Sorry, I think you meant to say, "philosophy of Science as it would be if performed by a Physicist." Philosophy of Physics, really. I suppose most of this is a completely normal byproduct of the fact that we're at the tail end of a century when physics was, as a discipline, technology, and economic activity, remarkably successful.

"West says the school inadequately prepares students for the black seas of infinity." --- I couldn't agree more.

A follow-up to that Gigerenzer and Hoffrage paper!

"In some sense the genomics companies were just too early for their own good (though the late entrants such as DeCode haven't fared much better). There are no genomics companies -- yet genomics is everywhere. Basic biology fueled by the genome or the technologies pushed by genomics permeate the drug industry (based on the 2 large pharmas I interviewed at in the year MLNM laid me off & what I can read; constructive dissent on this point is welcomed). Probably no novel small molecule drug development history will be directly pinned back to a 1990's genomics effort -- but also virtually no drugs going forward will have their development unaffected by the knowledge of the genome. Everything is tangled up & confused & merged." --- That's funny, some of the people from Millennium sit in Stata these days (in the Science Commons consortium). They have a similar perspective, I think, but a different takeaway message (also: more optimism).

Larry Wasserman's class notes from a course on regression analysis. Via cshalizi.

Wantwantwant. (purchasedpurchasedpurchased.)

I knew I'd seen this review somewhere before -- did I really not save it at the time (I guess not). I should probably go ahead and buy the book. "Although Sunstein correctly calls for “sound science” in risk policy, he often gets his science wrong and almost always attempts to reduce ethical to purely scientific questions." -- although that's a little question-begging itself. Then --- "Because most hazardous materials are not tested, most risk probabilities are determined through mathematical models. As such, the models describe events falling into the category of Bayesian “uncertainty,” where no accurate probabilities are available, because there are no frequency data. If data were available, there would be no need for risk analysis and its attendant models. Given this Bayesian uncertainty, virtually all risk experts accept the fact that risk analyses typically err by 4 to 6 orders of magnitude." -- What?

Standpipe?

Good enough that I would actually consider using it as a template for a future abstract.

"For some time now, I’ve been experimenting with algorithms for solving proportional analogies." Haven't had enough time to read the underlying paper yet (apparently, my family and girlfriend are not cool with me reading papers at the dinner table on Christmas Eve? who knew...), but this looks totally sweet. Time to add Apperceptual to the RSS reader (I'm not sure why I hadn't done that already).