Petter Häggholm

May 11, 21:21 UTC
Resolved - This issue has been resolved. Please trigger a new build by pushing a fresh commit. If you continue to see issues, please reach out to CircleCI support.

May 11, 21:13 UTC
Monitoring - We have identified the root cause of the issue. We have pushed a fix. Please trigger a new build by pushing a fresh commit.

May 11, 20:53 UTC
Investigating - Some customers using Dynamic Configuration may be seeing intermittent issues and are getting the following error "Max number of workflows exceeded."

Thank you for your patience as we continue to investigate.

According to Breaking Defense, the head of machine learning at Lyft, Craig Martell, has been named the Pentagon's Chief Digital and Artificial Intelligence Officer. From the report: The hiring of a Silicon Valley persona for the CDAO role is likely to be cheered by those in the defense community who have been calling for more technically-minded individuals to take leadership roles in the department. At the same time, Martell's lack of Pentagon experience -- he was a professor at the Naval Postgraduate School for over a decade studying AI for the military, but has never worked in the department's bureaucracy -- may pose challenges as he works to lead an office only months old. In an exclusive interview with Breaking Defense, Martell, who also worked as head of machine learning at Dropbox and led several AI teams at LinkedIn, acknowledged both the benefits and risks of bringing in someone with his background. [...] As CDAO, Martell will be responsible for scaling up DoD's data, analytics and AI to enable quicker and more accurate decision-making and will also play an important role in the Pentagon's Joint All-Domain Command and Control efforts to connect sensors and shooters. "If we're going to be successful in achieving the goals, if we're going to be successful in being competitive with China, we have to figure out where the best mission value can be found first and that's going to have to drive what we build, what we design, the policies we come up with," Martell said. "I just want to guard against making sure that we don't do this in a vacuum, but we do it with real mission goals, real mission objectives in mind." His first order of business? Figuring out what needs to be done, and how to best use the $600 million in fiscal year 2023 dollars the CDAO's office was marked for in the Pentagon's most recent budget request. "So whenever I tackle a problem, whenever I go into a new organization, the first questions that I ask are: Do we have the right people? Do we have the right processes? Do we have the right tools to solve the visions [and] goals?" Martell said. To tackle that, Martell wants to identify the office's "marquee customers" and figure out what's "broken in terms of... people, platform, processes and tools" -- a process that could take anywhere from three to six months, he added. "We really want to be customer-driven here," Martell said. "We don't want to walk in and say if we build it, they'll come."

Read more of this story at Slashdot.

"A federal judge in Washington blocked the Defense Department from restricting U.S. investment in the Chinese smartphone manufacturer Xiaomi Corp," reports Bloomberg: In the final days of the Trump administration, the Defense Department placed Xiaomi on a list of companies with alleged links to the Chinese military, triggering financial restrictions that were scheduled to go into effect next week. But on Friday, U.S. District Judge Rudolph Contreras put a temporary halt to the ban, siding with Xiaomi in a lawsuit that argued that the move was "arbitrary and capricious" and deprived the company of its due process rights. Contreras said Xiaomi was likely to win a full reversal of the ban as the litigation unfolds and issued an initial injunction to prevent the company from suffering "irreparable harm." After the ban was announced, the smartphone manufacturer faced the prospect of being de-listed from U.S. exchanges and deleted from global benchmark indexes. Xiaomi is the third-largest smartphone manufacturer in the world by volume. In the third quarter, it surpassed Apple Inc. in smartphone sales, according to the International Data Corporation.

Read more of this story at Slashdot.

His Invictus Games just started, but Prince Harry has been popping up all over Toronto, and with each stop, anticipation builds: Will his girlfriend, Meghan Markle, show up, too?

Some cats seem to recognize that a two-dimensional image of a bird on a t.v. or computer screen is not a real prey item, but others, like this cat, cannot.

It’s early in the morning in Queenstown, and I have a 7.5-hour bus ride to Fox Glacier ahead of me. All I can do this morning is direct your attention to a (free) New Yorker profile on Dan Dennett and his views on the mind, “Dan Dennett’s Science of the Soul.” I haven’t read it, but several readers brought it to my attention. His new book (which I have in Chicago but haven’t read either) is From Bacteria to Bach and Back: the Evolution of Minds; if you’ve read it, weigh in below.

Do note the use of the word “soul” in the tile; I would never have characterized Dan’s work as involving the “science of the soul.” But such is The New Yorker.

 

h/t: Nicole Reggia, John B.

 

One way to describe our overall editorial stance at SBM is that we are criticizing medical science in a constructive way because we would like to see higher standards more generally applied. Science is complex, medical science especially so because it deals with people who are complex and unique. Getting it right is hard and so we need to take a very careful and thoughtful approach. There are countless ways to get it wrong.

One way to get it wrong is to put too much faith in a new technology or scientific approach when there has not been enough time to adequately validate that approach. It’s tempting to think that the new idea or technology is going to revolutionize science or medicine, but history has taught us to be cautious. For instance, antioxidants, it turns out, are not going to cure a long list of diseases.

One recent technology that is very exciting, but insiders recognize is very problematic, is perhaps even more problematic than we thought –functional MRI scans (fMRI). A new study suggests that the statistical software used to analyse the raw data from fMRIs might be significantly flawed, producing a flood of false positive results.

An fMRI primer

MRI scanning uses powerful magnets to image soft tissue in the body. The magnets (1.5-3 Tesla, typically) align the spin of hydrogen atoms in water molecules with the magnetic field. The time it takes for the atoms to align and then relax depends on the characteristics of the tissue. The MRI scan therefore sees subtle differences in tissue (density, water content) and uses this information to construct detailed images.

Functional MRI takes it one step further – it images blood flow using a technique known as blood oxygenation level dependent (BOLD) imaging. When a part of the brain is active there is an initial dip in blood oxygenation as the neurons consume oxygen, but then blood flow increases bringing more oxygen to the tissue. This peaks at about 6 seconds, then the oxygen level dips below baseline and then back to baseline.

Software analyses the raw data from fMRI scan to create what are called voxels. A voxel is a three-dimensional pixel, representing a tiny cube of brain tissue containing about a million cells. The software determines the activity level of each voxel based upon how closely changes in oxygenation level match the expected pattern that occurs when brain cells are active.

The software will also look for groups of voxels that have the same activity level, a phenomenon known as clustering. Scientists infer from a clustering of voxels showing increased activity which in turn is inferred from how closely it matches the expected pattern of oxygenation, that a part of the brain is undergoing increased activity. Researchers try to correlate such regional activity with specific tasks, and then infer that this part of the brain is involved in that task.

The Black Box

Already we can see that interpreting the results of an fMRI study involves a chain of inference. The chain gets longer, however, because often researchers will need to do multiple trials of multiple individuals and then compare their brain activity statistically.

For example, you might have one control group that is essentially doing nothing, while another group is looking at a specific image. Then you compare the fMRI activity statistically between the two groups, and can conclude that looking at the specific image correlates with increased activity in one part of the brain, and from that you infer that the active part of the brain is involved in visual processing.

To date about 40,000 such studies have been published, adding a tremendous amount to our knowledge of what different parts of the brain are doing and how they connect up and interact.

However, in addition to other problems with fMRI studies, a key link in this long chain of inference has always been a bit of a controversy – the statistical analysis of the raw data to detect clusters of voxels that are active together. Researchers use statistical software packages that they purchase. They need to learn how to use the software, but most neuroscientists are not software engineers and so they have to trust that the software works as advertised. To them the software is a black box that spits out data.

Famously, neuroscientist Craig Bennett scanned a dead salmon using fMRI and found that when the dead salmon was shown pictures, the software detected a statistical cluster of voxel activity in its brain. Bennett won an Ig Nobel prize for his efforts. More importantly, he sounded an alarm bell that fMRI software might have a problem with generating false positives.

The new study

“Cluster failure: Why fMRI inferences for spatial extent have inflated false-positive rates,” was recently published in PNAS by authors Anders Eklund, Thomas E. Nichols, and Hans Knutsson. This is a follow up study to earlier work with similar results.

In their prior study they set out to conduct validation tests of one popular fMRI statistical software package. They used existing open source data for task-based in individual data, using a null data set (meaning there should be no difference). Given where the thresholds are set in terms of finding a statistical match to a pattern of oxygenation consistent with brain activity, there should be a 5% false positive rate. They found a false positive rate as high as 70%.

The current study expands on their earlier work. First, they are using three of the most common statistical software packages instead of just one, and they are also comparing groups rather than just individuals. They compare two groups each randomly drawn from the same pool of healthy individuals, so there should be no difference. Again, there should be a 5% false positive rate, but they found many statistical tests with much higher false positive rates, between 60-90%. (This depends on the parameters used, like number of subjects and threshold of statistical significance.) They summarize these results:

For a nominal familywise error rate of 5%, the parametric statistical methods are shown to be conservative for voxelwise inference and invalid for clusterwise inference.

There is a lot of statistical jargon in the paper, but basically the voxelwise inference tended to be conservative and have a valid false positive rate less than 5%, while the clusterwise inference tended to have high levels of false positive.

The authors also make two other important points. The first is that they corrected statistically for multiple comparisons when doing their analyses. (In other words, if you roll the dice 20 times, you are likely to get a 5% result once, so you have to adjust the statistical thresholds to account for the fact that you rolled 20 times.) However, in their review of published papers 40% do not correct for multiple comparisons. This means, in those 40% of papers, false positive rates are likely to be even higher.

They also complain that many researchers did not store their raw data in a way that would allow for a reanalysis of the data. We only have the results of the statistical analysis, but cannot run the analysis again with better methods or thresholds.

They also note that a specific software bug was discovered:

Second, a 15-year-old bug was found in 3dClustSim while testing the three software packages (the bug was fixed by the AFNI group as of May 2015, during preparation of this manuscript). The bug essentially reduced the size of the image searched for clusters, underestimating the severity of the multiplicity correction and overestimating significance (i.e., 3dClustSim FWE P values were too low).

What’s the fix?

This type of research is critical to the practice of science – it’s what makes science self-corrective. It also has to do directly with a concept that is critical to science but underappreciated, especially in the general public, that of validity.

Any test or measurement has to be validated, which usually means that the test is used with known data to see that it produces the results it should, that those results are meaningful, and that they are precise and reliable. Until a test has been validated, you don’t know what the results actually mean. They are like the personality tests given is Cosmopolitan Magazine; for entertainment purposes only.

What the authors of the current study are essentially saying is that the statistical software used to tease a signal out of the massive amount of noise generated by fMRI scans have not been properly validated. Prior validation studies were inadequate. Their studies are a better validation paradigm, and they show significant problems with the software that cause at least these three popular packages to generate high percentages of false positives.

The authors conclude:

It is not feasible to redo 40,000 fMRI studies, and lamentable archiving and data-sharing practices mean most could not be reanalyzed either. Considering that it is now possible to evaluate common statistical methods using real fMRI data, the fMRI community should, in our opinion, focus on validation of existing methods.

In other words, we have to dust ourselves off and just move forward. Let’s fix the statistical problems and make sure that fMRI studies going forward are more valid.

I have already had a skeptical eye toward fMRI research given all the crazy results that have been published (the dead salmon study aside). Now I am even more skeptical. Some of the research is clearly very rigorous and high quality, but much of it had many red flags for false positive results (sloppy methods, lack of replication, implausible results).

Hopefully this research and similar validation research will tighten up the field of fMRI research. It is an incredibly powerful tool, but it is tricky to use.

 

 

While I’m bashing the New York Times today, let me add this beef.

It amazes me how people can discuss the origins of ISIS-like brutality, and even allude to the influence of the “potent doctrine” that fuels and channels it, without mentioning religion itself. It’s part of the same mentality that makes Obama shy away from mentioning Islam as an influence on terrorism, or grudgingly recognize it by saying that ISIS is a “perversion of Islam.” The whole object is to avoid saying anything bad about any religion. To that my response is this: “If you say that religion can inspire good acts, why are you so reluctant to say that it can inspire bad ones?”

David Brooks is one of the stable of New York Times op-ed writers who refuses to mention the I- and M-words when discussing ISIS. In his new column “How ISIS makes radicals” (curiously, the original title was “How radicals are made”), Brooks manages to discuss the origins of violent Islamism while alluding to religion as a contributing factor, but avoiding all mention of religion except for one sentence in his thesis:

But the crucial issue, it seems to me, is what you might call the technology of persuasion — how is it that the Islamic State is able to radicalize a couple living in Redlands, Calif.?

Yes, the only mention of religion is the name “Islamic State.”

Brooks’s analysis of its origins draws heavily on Eric Hoffer’s famous book The True Believer: Thoughts on the Nature of Mass Movements, which I read in college. Brooks’s response is in fact Hoffer’s thesis: that mass movements are often a response to a dysfunctional society that impels disaffected youth to turn to a greater cause. But at least Brooks recognizes that ISIS and similar movements aren’t just the result of Western oppression, and in general what he says is sensible. Here are some excerpts (my emphasis):

. . . The purpose of an organization like ISIS is to get people to negate themselves for a larger cause.

Mass movements, [Hoffer] argues, only arise in certain conditions, when a once sturdy social structure is in a state of decay or disintegration. This is a pretty good description of parts of the Arab world. To a lesser degree it is a good description of isolated pockets of our own segmenting, individualized society, where some people find themselves totally cut off. [JAC: note that, as I’ve discussed often, increased religiosity is recognized by sociologists as a ubiquitous response to social dysfunction.]

The people who serve mass movements are not revolting against oppression. They are driven primarily by frustration. Their personal ambitions are unfulfilled. They have lost faith in their own abilities to realize their dreams. They sometimes live with an unrelieved boredom. Freedom aggravates their sense of frustration because they have no one to blame but themselves for their perceived mediocrity. Fanatics, the French philosopher Ernest Renan argued, fear liberty more than they fear persecution.

And here’s where religion begins to creep in, at least as I see it:

The successful mass movement tells such people that the cause of their frustration is outside themselves, and that the only way to alter their personal situation is to transform the world in some radical way.

To nurture this self-sacrificing attitude, the successful mass movement first denigrates the present. Its doctrine celebrates a glorious past and describes a utopian future, but the present is just an uninspiring pit. The golden future begins to seem more vivid and real than the present, and in this way the true believer begins to dissociate herself from the everyday facts of her life. . .

. . . Next mass movements denigrate the individual self. Everything that is unique about an individual is either criticized, forbidden or diminished. The individual’s identity is defined by the collective group identity, and fortified by a cultivated hatred for other groups.

What better way to take advantage of these disaffected youth than to lure them with the golden promise of religion, perhaps even a Caliphate? That is a cause far beyond oneself.  And here Brooks ventures solidly onto religious ground, but still refuses to recognize the territory:

These movements generate a lot of hatred. But ultimately, Hoffer argues, they are driven by a wild hope. They believe an imminent perfect future can be realized if they proceed recklessly to destroy the present. The glorious end times are just around the corner.

Glorious end times? Could that possibly be the Final Showdown Against Unbelievers that, says ISIS, will occur in the Middle East when the expanding Caliphate provokes the West?

And then Brooks quotes Hoffer in service of this thesis:

Hoffer summarizes his thought this way, “For men to plunge headlong into an undertaking of vast change they must be intensely discontented yet not destitute, and they must have the feeling that by the possession of some potent doctrine, infallible leader or some new technique they have access to a source of irresistible power. They must also have an extravagant conception of the prospects and potentialities of the future. Finally, they must be wholly ignorant of the difficulties involved in their vast undertaking. Experience is a handicap.”

Well, what is that “potent doctrine” but Islam? Who is the “infallible leader “but Abu Bakr Al-Baghdadi (channeling Muhammed)? What is the “prospect of the future” except the Caliphate?

At the end Brooks suggests the way to get rid of ISIS is to ameliorate the social dysfunction of its adherents, defeat them militarily, and “offer positive inspiring causes to replace the suicidal ones.” That sounds good, though it’s a long row to hoe. But how can we replace the “suicidal cause” unless we recognize what it is? Surely the prospect of Paradise is in there somewhere.

I’m not sure any of the NYT op-ed writers, including conservatives like Douthat, have ever explicitly mentioned that Islamic doctrine is a prime motivating force for ISIS. What a breath of fresh air it would be to see that said out loud!

Prosecutors from New York, London, Paris, and Madrid wrote an op-ed in yesterday's New York Times in favor of backdoors in cell phone encryption. There are a number of flaws in their argument, ranging from how easy it is to get data off an encrypted phone to the dangers of designing a backdoor in the first place, but all of that has been said before. And since anecdote can be more persuasive than data, the op-ed started with one:

In June, a father of six was shot dead on a Monday afternoon in Evanston, Ill., a suburb 10 miles north of Chicago. The Evanston police believe that the victim, Ray C. Owens, had also been robbed. There were no witnesses to his killing, and no surveillance footage either.

With a killer on the loose and few leads at their disposal, investigators in Cook County, which includes Evanston, were encouraged when they found two smartphones alongside the body of the deceased: an iPhone 6 running on Apple's iOS 8 operating system, and a Samsung Galaxy S6 Edge running on Google's Android operating system. Both devices were passcode protected.

You can guess the rest. A judge issued a warrant, but neither Apple nor Google could unlock the phones. "The homicide remains unsolved. The killer remains at large."

The Intercept researched the example, and it seems to be real. The phones belonged to the victim, and...

According to Commander Joseph Dugan of the Evanston Police Department, investigators were able to obtain records of the calls to and from the phones, but those records did not prove useful. By contrast, interviews with people who knew Owens suggested that he communicated mainly through text messages -- the kind that travel as encrypted data -- and had made plans to meet someone shortly before he was shot.

The information on his phone was not backed up automatically on Apple's servers -- apparently because he didn't use wi-fi, which backups require.

[...]

But Dugan also wasn't as quick to lay the blame solely on the encrypted phones. "I don't know if getting in there, getting the information, would solve the case," he said, "but it definitely would give us more investigative leads to follow up on."

This is the first actual example I've seen illustrating the value of a backdoor. Unlike the increasingly common example of an ISIL handler abroad communicating securely with a radicalized person in the US, it's an example where a backdoor might have helped. I say "might have," because the Galaxy S6 is not encrypted by default, which means the victim deliberately turned the encryption on. If the native smartphone encryption had been backdoored, we don't know if the victim would have turned it on nevertheless, or if he would have employed a different, non-backdoored, app.

The authors' other examples are much sloppier:

Between October and June, 74 iPhones running the iOS 8 operating system could not be accessed by investigators for the Manhattan district attorney's office -- despite judicial warrants to search the devices. The investigations that were disrupted include the attempted murder of three individuals, the repeated sexual abuse of a child, a continuing sex trafficking ring and numerous assaults and robberies.

[...]

In France, smartphone data was vital to the swift investigation of the Charlie Hebdo terrorist attacks in January, and the deadly attack on a gas facility at Saint-Quentin-Fallavier, near Lyon, in June. And on a daily basis, our agencies rely on evidence lawfully retrieved from smartphones to fight sex crimes, child abuse, cybercrime, robberies or homicides.

We've heard that 74 number before. It's over nine months, in an office that handles about 100,000 cases a year: less than 0.1% of the time. Details about those cases would be useful, so we can determine if encryption was just an impediment to investigation, or resulted in a criminal going free. The government needs to do a better job of presenting empirical data to support its case for backdoors. That they're unable to do so suggests very strongly that an empirical analysis wouldn't favor the government's case.

As to the Charlie Hebdo case, it's not clear how much of that vital smartphone data was actual data, and how much of it was unable-to-be-encrypted metadata. I am reminded of the examples that then-FBI-Director Louis Freeh would give during the First Crypto Wars in the 1990s. The big one used to illustrate the dangers of encryption was Mafia boss John Gotti. But the surveillance that convicted him was a room bug, not a wiretap. Given that the examples from FBI Director James Comey's "going dark" speech last year were bogus, skepticism in the face of anecdote seems prudent.

So much of this "going dark" versus the "golden age of surveillance" debate depends on where you start from. Referring to that first Evanston example and the inability to get evidence from the victim's phones, the op-ed authors write: "Until very recently, this situation would not have occurred." That's utter nonsense. From the beginning of time until very recently, this was the only situation that could have occurred. Objects in the vicinity of an event were largely mute about the past. Few things, save for eyewitnesses, could ever reach back in time and produce evidence. Even 15 years ago, the victim's cell phone would have had no evidence on it that couldn't have been obtained elsewhere, and that's if the victim had been carrying a cell phone at all.

For most of human history, surveillance has been expensive. Over the last couple of decades, it has become incredibly cheap and almost ubiquitous. That a few bits and pieces are becoming expensive again isn't a cause for alarm.

This essay originally appeared on Lawfare.

EDITED TO ADD (8/13): Excellent parody/commentary: "When Curtains Block Justice."

In general people think they know more than they do. This is arguably worse than mere ignorance – having the illusion of knowledge. Psychologist David Dunning (of the Dunning-Kruger effect) recently wrote in an editorial about his own study (which I discuss here):

“What’s curious is that, in many cases, incompetence does not leave people disoriented, perplexed, or cautious. Instead, the incompetent are often blessed with an inappropriate confidence, buoyed by something that feels to them like knowledge.”

Dunning was discussing the “ignorant mind.” Further, self-perceived expertise does not protect against this effect and in fact may make it worse. A new paper published in the latest issue of Psychological Science, When Knowledge Knows No Bounds: Self-Perceived Expertise Predicts Claims of Impossible Knowledge, presents four studies exploring the relationship between perceived expertise and the illusion of knowledge.

David Dunning is the last author on this paper as well, along with Stav Atir and Emily Rosenzweig. They present the results of four studies. The first establishes the basic phenomenon: self-perceived expertise in finance predicts claiming knowledge in nonexistent financial concepts (pre-rated stocks, fixed-rate deduction, annualized credit – they did a good job making up fake technobabble.) This effect is called “overclaiming.”

That is a good research model, because there is no way someone could have actual knowledge about something that does not exist (even as a false idea, because it was made up for the study itself).

In the second study they found that overclaiming within a specific field correlates with increased perceived knowledge within that field, more so than knowledge in general. This suggests that it is the perception of knowledge about a specific topic that is driving overclaiming about that topic, not a more general tendency to overclaim.

In the third study they addressed the possibility that overclaiming results from “impression management.” In other words, in order to maintain the impression that one is an expert in biology, one will claim that they have specific knowledge about made-up biological terms. Therefore they have insight into the fact that they do not really know the made-up term but are lying to maintain their image as an expert.

To test this hypothesis they warned the third group that some of the terms in the list were made up. The idea is that if overclaiming were driven entirely or mostly by impression management, fear of claiming to know a made up term would mitigate the effect. It did not, however, suggesting that impression management is not driving overclaiming.

To test further the alternate hypothesis that overclaiming is driven by the perception of expertise itself they did a fourth study in which they tried to manipulate the perceived knowledge of the subjects. They divided them into three groups and gave group one an easy quiz on geology, group two a difficult quiz, and group three no quiz. The idea here is that taking an easy quiz and doing well will make one feel like more of an expert.

They found that the group that took the easy quiz on geology had an increased perceived familiarity with nonexistent locations.

Taken together these studies suggest that the perception itself that someone has expertise in a specific topic leads to the illusion of knowledge in that topic. In these series of studies they were measuring the perception of knowledge, and not actual knowledge.

These results add to Dunning’s previous work. The Dunning-Kruger effect demonstrated that the worse an individual did on tests of humor, grammar, and logic the more they overestimated their performance. This series of studies suggest that ignorance robs an individual of the ability to assess their own level of knowledge.

As I discussed previously, Dunning later commented that this ignorance is often not the absence of knowledge but the illusion of knowledge, driven by “the clutter of irrelevant or misleading life experiences, theories, facts, intuitions, strategies, algorithms, heuristics, metaphors, and hunches.”

These new studies extend this work, indicating that the illusion of knowledge does not only increase one’s self-assessment of knowledge, but also causes overclaiming – pretending to have knowledge about topics of which one is actually ignorant.

The authors of the new paper also speculate that the illusion of knowledge in a particular area may cause someone to feel that they do not have to further study that area of knowledge, leading to ignorance of the exact area in which they claim expertise. That sounds like an interesting area of further research.

As always, I encourage my readers to apply these lessons not only to others but to themselves. The Dunning-Kruger effect and the illusion of knowledge apply to everyone, not just to others.

Opterator is a decorator to make quick and clean command line interfaces for scripts. It’s not designed to compete with argparse or click. Instead it supports that common usecase where you’re writing a reusable script and don’t want to write all the bits required for argument parsing, but also don’t want to use the maligned ‘just access sys.argv and pray’ technique.

Say you’ve written a script that makes a thumbnail of an image. It requires the image name, the output thumbnail name, and an optional algorithm. Just do this:

from opterator import opterate

@opterate
def main(image, thumbnail, algorithm="cubic"):
    # do the scaling stuff

if __name__ == "__main__":
    main()

There’s no mucking about with options, and you even get a helptext that is borderline useful:

python3 meh.py -h
usage: meh.py [-h] [-a ALGORITHM] image thumbnail

positional arguments:
  image
  thumbnail

optional arguments:
  -h, --help            show this help message and exit
  -a ALGORITHM, --algorithm ALGORITHM

If that’s not good enough, you can customize the helptext by editing the docstring, as described in the
README.

But opterator has supported that for years. Let’s talk about some new stuff. I recently released version 0.5, which has a few bugfixes and a couple exciting new features that I want to introduce.

First, I ported it from optparse to argparse. This makes opterator handle positional arguments much more neatly, both in the code and in the resulting helptext. It doesn’t really change your interaction with opterator as an api client, but the users will see improved helptext.

Second, I’ve added support for function annotations so you can change the variable name output by opterator. This obviously only works on Python 3, since function annotations are a new feature in Python. Opterator itself, however, is Python 2 and 3 compatible as it has been since 2012. Here’s the same example, but we changed the algorithm argument to be passed in as –scaling or -s:

from opterator import opterate

@opterate
def main(image, thumbnail, algorithm:["--scaling", "-s"]="cubic"):
    # do the scaling stuff
    pass

if __name__ == "__main__":
    main()

Here’s the resulting helptext:

usage: meh.py [-h] [--scaling ALGORITHM] image thumbnail

positional arguments:
  image
  thumbnail

optional arguments:
  -h, --help            show this help message and exit
  --scaling ALGORITHM, -s ALGORITHM

The main reason you might want to do this is that you’ve got options that start with the same letter (the first letter is what opterator uses as a short option, by default). Function annotations let you supply other options if you like. Or possibly you just have a burning desire to play with function annotations. Either way, opterator has you covered.

Opterator is well-tested, so I’ve (belatedly) upgraded the classifier from beta to production. It’s a small tool that has saved me a lot of time and many lines of ugly code over the years. I think it’s worth having installed in site- packages so you can write quick scripts without thinking about getting data from the user.

God damn it woman, pay attention.

Bowe Bergdahl, the U.S. Army sergeant who has been recovering in Germany after five years as a Taliban captive, returned to his home country early Friday to continue his medical treatment.

From HuffPo via reader Don, we have an amazing video of a crow solving a complex puzzle. We all knew that corvids were smart, but did you know that they were this smart?

Here are the HuffPo notes:

In this BBC special, Dr. Alex Taylor has set up an eight-step puzzle to try and stump one of the smartest crows he’s seen in captivity. They describe the puzzle as “one of the most complex tests of the animal mind ever.”

This isn’t the first time crows’ intelligence has been tested, either. Along with being problem solvers, these animals have an eerie tendency towards complex human-like memory skills. Through several different studies, we’ve learned that crows can recognize faces, communicate details of an event to each other and even avoid places they recognize as dangerous.

This bird, dubbed “007″ for its crafty mind, flies into the caged puzzle and spends only seconds analyzing the puzzle before getting down to business. Despite the puzzle’s difficulty, the bird only seems to be stumped momentarily. At the end of the puzzle is a food reward, but how he gets there is what will really blow your mind.

NOTE: If the video doesn’t play, or you get a video that’s not about crow intelligence, watch it directly on the HuffPo page.

Now I don’t know how much the bird was trained beforehand, and whether it knew, for instance, that putting three stones in the lucite box would release a longer stick, but I doubt it. If the bird figured all that out de novo, it’s truly remarkable.  But since the bird apparently hadn’t had experience with both the stones and the small stick it used retrieve them, or with the stones and the box that released the large stick (the narrator says that the bird had experience only with single items of the 8-item puzzle), then it is a stunning demonstration of bird intelligence.

We know that the 100 trillion microbes in the human body are important to our health. What’s harder to know is how to use them to make us healthy.

Normally, our resident microbes–the microbiome–carry out a number of important jobs for us, from fighting off pathogens to breaking down food for us. If they get disrupted, we  suffer the consequences. Sometimes antibiotics can upset the ecological balance in our bodies so severely, for example, that rare, dangerous species can take over.

For decades, doctors and scientists have searched for microbes that can promote our health by taking up residence in our bodies. They’ve had some modest success in treating people by giving them a single species at a time. (Don’t be fooled by all the so-called probiotic foods and pills you can buy over the counter–few, if any, have ever been scientifically shown to be effective.) Part of the problem has been that scientists haven’t been terribly systematic about searching for microbes. Very often, their most important standard for a probiotic germ is not its healing power, but its ability to survive in food, or packed into a pill.

Scientists have also had some limited success at the other end of the spectrum, by exploiting much of the microbiome’s diversity at all once. For some people with deadly gut infections, for example, the only cure is to get a so-called fecal transplant from a healthy donor. Fecal transplants show a lot of promise, but scientists don’t have a clear idea of how they work. A stool sample is loaded with hundreds of different species, any one of which might be either essential for overthrowing pathogens or just along for the ride.

In Nature today, a team of scientists report taking an important step forward in microbiome-based medicine. They searched for species methodically, in the same way medicinal chemists search for new drugs. They have pinpointed a handful of species living in the human gut that collectively show signs of fighting effectively against some autoimmune diseases.

Scientists have long suspected that the immune system would benefit from microbiome-based medicine. That’s because immune systems depend on the microbiome to develop normally in the first place. As a child’s immune cells grow and divide, they pick up signals from the microbes. Those signals teach the immune system to become tolerant. They can still recognize a dangerous pathogen and kill it, but they spare the beneficial bugs. And they also become less likely to overreact to a harmless molecule or our own tissues.

One important group of immune cell involved in this tolerance are known as Tregs (short for the CD4+ FOXP3+ T regulatory cells). They are abundant in the microbe-packed gut, where they help to broker a truce between host and the germs that live there. Tregs also depend on the microbiome for their very existence. When scientists rear mice without any germs at all, the animals develop very few Tregs. And as a result their immune system becomes prone to raging out of control.

A number of experiments have hinted that abnormal levels of Tregs are behind some autoimmune gut diseases, such as colitis, which causes chronic inflammation in the large intestines and diarrhea. This raises the possibility that a treatment for colitis would be to bring Tregs back to normal levels. Perhaps among all the bacteria in the gut, scientists could find the ones that sent the signals to the Tregs.

Over the past few years, Kenya Honda of the University of Tokyo and his colleagues have been hunting for those species. They started by testing out subsets of the microbiome to find groups of species that could foster the growth of Tregs. Raising healthy mice, they collected the mouse droppings, which contained lots of bacteria. They treated the droppings with chloroform, which kills most bacteria. The only microbes that survived were species that make spores tough enough to withstand the chemical.

When the scientists gave those spore-forming bacteria to germ-free mice, the level of Tregs in the animals went up. That didn’t happen when the scientists gave the mice other kinds of bacteria instead.

This discovery didn’t pinpoint exactly which bacteria were fostering Tregs. But it certainly narrowed down the line-up of suspects dramatically. And it also prompted Honda and his colleagues to start acting more like a drug company testing promising new compounds. In fact, Honda co-founded a company called Vedanta Biosciences for that express purpose.

Their goal now became finding a species, or a group of species, that live in humans, and which promote the growth of Tregs in the gut. Rather than taking the kitchen-sink approach of fecal transplants, they would try to deliver a surgically precise germ.

They took stool samples from a healthy Japanese volunteer and doused them with chloroform, so that once more they only had to contend with spore-forming species. Then they inoculated germ-free mice with different combinations of the surviving bacteria. After letting the bacteria grow in the mice, they then inspected the animals to see if any of them had high levels of Tregs as a result. They did find some of those combinations, and they went on narrowing down the suspects until  they were left with just seventeen species, along belong to a type of bacteria called Clostridia.

One particularly interesting result of the experiment was that they couldn’t get these good results from any fewer than those seventeen species. On its own, each of the species was unable to foster the immune cells. It may be a combination of signals from all seventeen species that promotes the Tregs.

The researchers wondered if these seventeen species played a special role in autoimmune diseases in humans. To find out, they looked at the microbiomes from health people and people with colitis. The seventeen species tended to be rarer in the sick people.  Perhaps, the researchers reasoned, losing this network of microbes weakens the signals that keep Treg levels normal. The immune system spins out of control, leading to colitis.

If that were true, then giving someone a pill with all seventeen species might be an effective way to fight colitis. But Honda and his colleagues weren’t ready to start inoculating people with Clostridia. Clostridia is a huge group of species that includes some very nasty characters that cause diseases like tetanus and botulism. They would have to do some preliminary work first.

All seventeen species were new to science (something that’s pretty typical for microbiome research), so the scientists sequenced their genomes. None of the seventeen species carried genes for toxins or other disease-causing proteins. From an inspection of their DNA, at least, the microbes looked safe.

Next, the scientists tested the bacteria out on mice. They gave the microbes to animals either suffering from colitis or from allergy-triggered diarrhea. In both cases, the bacteria raised the level of Tregs dramatically in the guts of the mice. The mice also partly recovered from their diseases. The mice with colitis had less inflammation, and the mice with diarrhea had healthier stool.

From here, the Vedanta researchers eventually hope to get to clinical trials on humans. As I wrote in April, turning bugs into drugs is a big challenge on many levels. For one thing, the FDA doesn’t have a long tradition of approving such research. And while Honda and his colleagues have certainly gone a long way to pinpointing how microbes foster Tregs, they have yet to work out the precise balance of signals that really matters to the immune system.

Nevertheless, a seventeen-bug cocktail would be appealing in many ways. For one thing, the microbes are regular residents of the human gut, dwelling there for people’s entire lifetime. And they only live in the gut, and not the heart or the liver or some other organ. Both these facts suggest that such a cocktail would be unlikely to cause harmful side effects. What’s more, the microbes would be able to deliver a steady, long-running dose of the chemicals necessary to keep Treg levels healthy.

The new study also points to a way to systematically search the microbiome for treatments for other diseases. It’s possible that small teams of other species handle other jobs in the body. They may nurture other types of immune cells, for example. Or they may send signals into the body that regulate body weight. By winnowing down the microbiome, scientists may be able to deploy those elite units to fight other diseases.

[Update: Nature paper link fixed]

Oh, those wacky mixed-up Christians and their warped authoritarianism. There’s this…thing… going around called Christian Domestic Discipline (CDD), in which Heads of Households (HoHs; there are lots of acronyms here) are encouraged to spank their wives. That’s right, it’s always husbands doing the spanking — anything else would violate God’s natural order, of course. Christian Domestic Discipline, or CDD as its adherents call it, is a movement that seeks to carry out God’s will. Which specific plan of God’s? Oh, you know, just that all women obey their husbands fastidiously — a dynamic that CDD thinks is best maintained through doling out out corporal punishments. Its few thousand practitioners, however, claim that it’s not domestic abuse. The very-serious practitioners have this discipline have conveniently put together a 50 page guide to spanking your wife (pdf) — somehow these obsessive cranks always get carried away trying to justify their abuses. It’s terrible and ridiculous. Read the thing, and all you learn is how much these kooks want to infantilize women.

You can now re-visit the first web page ever — built 21 years ago — by directing your browser to the first ever web address.

by Matthew Cobb

Crispian Jago has a blog (not a website) called The Reason Stick. The other day he posted this excellent Venn Diagram of Irrational Nonsense., with four fields: Religious Bollocks, Quackery Bollocks, Pseudoscientific Bollocks and Paranormal Bollocks. At its heart, like a malevolent spider (sorry, spiders!) sits Scientology… Click to enbiggen and find your favourite woo!

h/t @SLSingh