I bought a pair, hoping to explore using it to keep notes for my slides and help track time when I'm doing a presentation and the like.
I bought it right after Facebook did the Oculus Rift acquisition, when I canceled my dev-kit order, and I wanted a thing I could fiddle around for development purposes.
So far in exchange for my trouble, I mostly get to stop and answer questions about Google Glass several times a day when I wear them. That much isn't so bad.
Now I have a device I wear that has to maintain a constant link to my phone, draining its battery, so now I have to recharge two devices faster and I can't use it as 'more convenient' navigation without getting out my phone anyway to go to the app to turn on GPS, so its day-to-day usage is just flat-out painful.
Oh, and I have to carry an extra pair of glasses, despite having switched the Glass to prescription lenses.
Why? If I walk to work, which takes about an hour and a half, if I use the glass at all during the trip, it is typically out of juice by the end of the walk, so now I have to plug my glasses in at the office, which means I need to get out another pair so I can still see.
And I better remember to carry the case, because if I go to the movie, the MPAA will get me arrested if I forget and wear them in, but since they don't fold up, I have to choose between a huge hard case or a big bulky pouch I'm constantly worried will go crunch.
Oh, and I'd better switch to my real glasses when I drive, lest I get arrested for that, too.
Oh, and if I walk by a school I get paranoid parents who think I'm out to take candid shots of their precious children, despite having a third party lens cap on.
I've had some punk kid try to rip them off my face and run on the T, so there is an apparently increased theft risk.
Now, because they polarize the glass in the prism they use to reflect light to your eye you can't get the lenses polarized or treated with any sort of anti-glare, but if you walk around in sunlight light reflects off the bottom of the prism into your eye constantly.
There is a little bit of silver mirroring that is just deposited on the end of the prism -- not covered with anything. I went for a walk in Australia on a humid, high UV day. It just flaked off, which effectively dropped my screen to about 10% brightness. They did replace it, but it meant a few weeks without a device, during which I decided I didn't really miss the inconvenience.
So in exchange for $1700 or so (after adding prescription lenses) I get to get called a glasshole by the internet and get treated as a even evil child-stalker road-hazard pirate pariah by society, and have to carry another pair of glasses anyways.
My perception of a pristine and peaceful scuba dive changed when I went for my first dive off the Great Barrier Reef. It looks so easy in the movies, but the breathing underwater with an oxygen mask is difficult. Don’t let those practice lessons in the swimming pool fool you! To make it easier for us, here is the Triton Oxygen Mask For Diving. It is a very convenient oxygen respirator concept that allows us to breathe under water for a long time by simply biting it. It also does not require the skill of breathing in and out while biting mouthpiece like conventional respirator.
Triton uses a new technology of artificial gill model. It extracts oxygen under water through a filter in the form of fine threads with holes smaller than water molecules. This is a technology developed by a Korean scientist that allows us to freely breathe under water for a long time. Using a very small but powerful micro compressor, it compresses oxygen and stores the extracted oxygen in storage tank. The micro compressor operates through micro battery. The micro battery is a next-generation technology with a size 30 times smaller than current battery that can quickly charge 1,000 times faster. Triton is a 2013 sadi product innovation studio project.
There are no real images of this "product" as it is a design concept, not reality. No science is behind this design, just magic and wishful thinking. It is pretty clear that the distinction has been lost in translation.
It simply isn't big enough to supply enough oxygen to work.
The problem here is the mean abs dev for a larger dataset can't be composed out of mean abs deviations for the smaller subsets. e.g. Having the mean abs devs for a bunch of days is useless for computing it for the month.
It just isn't a very good number to use for aggregations in that it is less useful than median absolute deviation for characterizing the data robustly and dramatically harder to compute than standard deviation.
The article naively makes it out like standard deviation is some huge historical mistake, while holding out a mediocre tool as a universal replacement.
Amusingly had he argued for the even harder to calculate median absolute deviation he'd have had a much stronger case, as it fits his overall "antifragile" narrative much better than mean abs dev in some senses and is even smaller, which he seems to like in the article for some reason.
Doing two passes means I cannot update a previous answer in response to small deltas of information. If I have 2 petabytes of data and 20 megs of updates come in, I have to go back to the well and touch all 2 petabytes, I can't just update my sufficient statistics. The incompressibility of this statistic was what I was referring to with that comment. It rules out many scenarios.
If I am interested in just moments around the mean I can collect them as summary statistics in tree form, and usually compute them in log time. That use case is also gone. At the scale of data I routinely work with O(n) isn't in the cards and the incompressibility of either MAD works actively against it more so than the 2x factor would naively make it sound,
There exist approximations to the mean abs dev that can be computed online that converge in the limit, but nothing that yields the correct answer exactly.
Mean absolute deviation is a useful statistic, though I tend to actually prefer median absolute deviation.
You can actually prove that.
median abs dev = mean abs dev = standard deviation
You can also prove that median abs dev will provide the minimal absolute deviation from any number, so in that sense mean abs dev is kind of a strange choice here.
That said, we can say a few things about it.
It is a pain in the ass to calculate. It also tends to favor solutions that let outliers run wild. Least squares provides a compromise that favors explaining the outliers over fiddling around trying to slightly better fit the rest, but doesn't go so far as to give all the weight to the outliers like higher norms.
Often you can actually start from something based on L2 norm and 'improve it' to L1 norm, e.g. this is commonly done to refine a solution in terms of least squares in terms of least absolute deviation.
Standard deviation is mostly used to get a feel for the 'shape' of how spread out your data is.
I'm not saying mean abs dev it isn't a useful statistic, but like most things promoted by Nassim Taleb it isn't the panacea he purports it to be, and on sufficiently large data sets where calculating a proper median abs dev can be prohibitively expensive or nigh impossible, the mean abs dev isn't going to be enough "better" than the standard deviation for most purposes that it is worth the loss of ability to compute it online.
Sadly, the problem is your proposed algorithm doesn't work.
Consider the sequence [1,2..100], The real mean absolute deviation is 25.
Your algorithms yields er.. something around 1?
Mean absolute deviation requires you to sum over a bunch of absolute values of differences to a number you don't know a priori.
Unlike stddev's use of x^2, abs doesn't have a continuous derivative and can't be split out into calculations in terms of the moments around 0, and you can't borrow Chan's algorithm.
Basically, to my knowledge, there isn't a sufficient statistic you can accumulate for MAD (either version of MAD) that takes less space than the original data.
On the other hand, you also need to use 2-pass algorithms to compute Mean Absolute Deviation, whereas STD can be easily calculated in one pass. And you still need standard deviation as it relates directly to the second moment about the mean.
Also, annoyingly, Median Absolute Deviation competes for the MAD name and is more robust against outliers.
Also, on github you can still tag versions and it'll still put up tar balls you just can't upload some zip file or your own and get it in the same place as the tagged versions.
At a convention the people aren't usually sitting at their computers the whole time huddled over problems they are working on. A hackathon for me is a group setting of people working on code. A convention has them in breakout sessions talking about it, but not writing it.
I'm going to disagree with most of the replies I've seen here so far about just piling on constraints and limitations.
When I go to a hackathon, I am looking for an open forum with interesting people to talk to and people who have their own problems to solve. I get sucked into new problems for 2-3 days and I emerge on the other side with insights into areas I wouldn't have thought of working on before.
I'm not looking for structure from the organizer about what to work on. I and most of the people I know already have a ton of projects in the wings. I'm looking for a good collaborative space to talk to people, people who've brought interesting projects to help, and whiteboard/blackboard space to use for explaining things.
The Haskell hackathons (Hac-Phi and Hac-Boston in particular) have generally followed this format and I love them.
I've gone to other events where someone is trying to harness a hackathon to achieve some particular end and pass out prizes or something, and in general I've been bored out of my mind. If I want to go work with some fixed group of people on some fixed task I can do that. It is called a job.
I'm at a hackathon to generally improve the state of things that the people around me are passionate about and to be exposed to new things.
As a CIO, I viewed my job to be the opposite of everything in this article.
Of course it is good to listen. It is good to be able to interact with anyone on their level of technical expertise and understanding. This advice holds at every level of an organization.
It is also occasionally good to be capable of being demonstrably the most technically competent person in the room. Effective organizations do need the person who can actually ensure there exists an implementble strategy to accomplish the things the CEO is selling the world, and the things the client wants, and who can articulate to vendors exactly why their magic bullet isn't quite what you need. And in many ways as a CIO, your role is to be the one person at that level of management who really understands the ins and outs of how the technology works, how things can improve and how you can adapt to meet the challenges of the organization as a whole.
Sometimes that means being the voice of reason as the curmudgeonly technology guy, but more often it means trying to steer management towards implementable solutions and being able to suggest things that give the other CXO types options they didn't know existed.
Whether facing inward within the organization or outward to clientele or vendors, you need to be able to communicate effectively. One thing this article omits is that when facing outward, it is often good to know when to overload the vendor to get to someone who is more competent to address your concerns, and somewhat more judiciously to be able to out-tech a client's technical guys as well.
Sometimes it _does_ pay to be the smartest person in the room.
They could probably do even better by getting the phone to emit a loud clapping sound, approximating a dirac delta so they could measure the impulse response of the room. The fourier transform of that should have some nice distinctive shapes. On the other hand, that wouldn't be nearly as unobtrusive, and most smart phones have crappy speakers, so you wouldn't get much response.
I used CLEPs for 30 credits of a double B.S. including almost all of my general education requirements. The key with getting most universities to accept CLEPs -- which are mostly targeted at people coming out of highschool with some college level experience -- is to make sure that you haven't taken an official college level course in the department that you plan to CLEP in. I CLEP'd out of literature, american history, sociology, american government, composition, economics... Basically everything I could. Take AP exams for everything you can as well. I still wound up having to take a couple of classes outside of math and computer science in order to fulfill art and cross-cultural coursework requirements -- the one that stuck out most was african american theatre -- but overall I can't say I'm any worse off for having given the other departments their pound of flesh.
I did have to study for the CLEPs I took to ensure I passed them all, but I was much happier studying sociology for a weekend than for a semester.
I stand corrected. I had at first thought general relativity was being used. I should have RTFA.
That said, while string theory can unify them on paper, string theory comes off as more of a religion than a science given that most of its flavors are completely non-falsifiable, because they have so many degrees of freedom that they can excuse pretty much anything.
A left handed dvorak user maxes out at 6 characters with something like 'papaya' or 'yuppie.';)
Even a right handed qwertyist at least gets to 9 with 'polyphony.'
By contrast a right handed Dvorak user can't even put the . on the abbreviation for boulevard 'blvd.' Thats what you get for putting all of your eggs (er vowels) in one basket (er hand) and sliding your punctuation around.;)
In other news, Blizzard is looking forward to record profits in the coming quarters from windfall lawsuits against the government willfully violating their terms of service.
In other other news, after reaching an agreement with the US government, the Chinese government seeks to reach similar terms to investigate Turing-complete gold farmers pointing out that gold farming is an intrinsically human activity and that a bot that could accomplish the same ends would be a great boon to society.
Slashdot Mirror
I bought a pair, hoping to explore using it to keep notes for my slides and help track time when I'm doing a presentation and the like.
I bought it right after Facebook did the Oculus Rift acquisition, when I canceled my dev-kit order, and I wanted a thing I could fiddle around for development purposes.
So far in exchange for my trouble, I mostly get to stop and answer questions about Google Glass several times a day when I wear them. That much isn't so bad.
Now I have a device I wear that has to maintain a constant link to my phone, draining its battery, so now I have to recharge two devices faster and I can't use it as 'more convenient' navigation without getting out my phone anyway to go to the app to turn on GPS, so its day-to-day usage is just flat-out painful.
Oh, and I have to carry an extra pair of glasses, despite having switched the Glass to prescription lenses.
Why? If I walk to work, which takes about an hour and a half, if I use the glass at all during the trip, it is typically out of juice by the end of the walk, so now I have to plug my glasses in at the office, which means I need to get out another pair so I can still see.
And I better remember to carry the case, because if I go to the movie, the MPAA will get me arrested if I forget and wear them in, but since they don't fold up, I have to choose between a huge hard case or a big bulky pouch I'm constantly worried will go crunch.
Oh, and I'd better switch to my real glasses when I drive, lest I get arrested for that, too.
Oh, and if I walk by a school I get paranoid parents who think I'm out to take candid shots of their precious children, despite having a third party lens cap on.
I've had some punk kid try to rip them off my face and run on the T, so there is an apparently increased theft risk.
Now, because they polarize the glass in the prism they use to reflect light to your eye you can't get the lenses polarized or treated with any sort of anti-glare, but if you walk around in sunlight light reflects off the bottom of the prism into your eye constantly.
There is a little bit of silver mirroring that is just deposited on the end of the prism -- not covered with anything. I went for a walk in Australia on a humid, high UV day. It just flaked off, which effectively dropped my screen to about 10% brightness. They did replace it, but it meant a few weeks without a device, during which I decided I didn't really miss the inconvenience.
So in exchange for $1700 or so (after adding prescription lenses) I get to get called a glasshole by the internet and get treated as a even evil child-stalker road-hazard pirate pariah by society, and have to carry another pair of glasses anyways.
There are no real images of this "product" as it is a design concept, not reality. No science is behind this design, just magic and wishful thinking. It is pretty clear that the distinction has been lost in translation.
It simply isn't big enough to supply enough oxygen to work.
The problem here is the mean abs dev for a larger dataset can't be composed out of mean abs deviations for the smaller subsets. e.g. Having the mean abs devs for a bunch of days is useless for computing it for the month.
It just isn't a very good number to use for aggregations in that it is less useful than median absolute deviation for characterizing the data robustly and dramatically harder to compute than standard deviation.
The article naively makes it out like standard deviation is some huge historical mistake, while holding out a mediocre tool as a universal replacement.
Amusingly had he argued for the even harder to calculate median absolute deviation he'd have had a much stronger case, as it fits his overall "antifragile" narrative much better than mean abs dev in some senses and is even smaller, which he seems to like in the article for some reason.
Those were less than or equals that got eaten by slashdot. Bah.
median abs dev is less than or equal to mean abs dev which is less than or equal to standard deviation.
So while Nassim is correct that mean abs dev is always smaller than standard deviation, the median abs dev is smaller still.
Doing two passes means I cannot update a previous answer in response to small deltas of information. If I have 2 petabytes of data and 20 megs of updates come in, I have to go back to the well and touch all 2 petabytes, I can't just update my sufficient statistics. The incompressibility of this statistic was what I was referring to with that comment. It rules out many scenarios.
If I am interested in just moments around the mean I can collect them as summary statistics in tree form, and usually compute them in log time. That use case is also gone. At the scale of data I routinely work with O(n) isn't in the cards and the incompressibility of either MAD works actively against it more so than the 2x factor would naively make it sound,
There exist approximations to the mean abs dev that can be computed online that converge in the limit, but nothing that yields the correct answer exactly.
I am unaware of such an algorithm. Reference?
Mean absolute deviation is a useful statistic, though I tend to actually prefer median absolute deviation.
You can actually prove that.
median abs dev = mean abs dev = standard deviation
You can also prove that median abs dev will provide the minimal absolute deviation from any number, so in that sense mean abs dev is kind of a strange choice here.
That said, we can say a few things about it.
It is a pain in the ass to calculate. It also tends to favor solutions that let outliers run wild. Least squares provides a compromise that favors explaining the outliers over fiddling around trying to slightly better fit the rest, but doesn't go so far as to give all the weight to the outliers like higher norms.
Often you can actually start from something based on L2 norm and 'improve it' to L1 norm, e.g. this is commonly done to refine a solution in terms of least squares in terms of least absolute deviation.
Standard deviation is mostly used to get a feel for the 'shape' of how spread out your data is.
I'm not saying mean abs dev it isn't a useful statistic, but like most things promoted by Nassim Taleb it isn't the panacea he purports it to be, and on sufficiently large data sets where calculating a proper median abs dev can be prohibitively expensive or nigh impossible, the mean abs dev isn't going to be enough "better" than the standard deviation for most purposes that it is worth the loss of ability to compute it online.
Sadly, the problem is your proposed algorithm doesn't work.
Consider the sequence [1,2..100], The real mean absolute deviation is 25.
Your algorithms yields er.. something around 1?
Mean absolute deviation requires you to sum over a bunch of absolute values of differences to a number you don't know a priori.
Unlike stddev's use of x^2, abs doesn't have a continuous derivative and can't be split out into calculations in terms of the moments around 0, and you can't borrow Chan's algorithm.
Basically, to my knowledge, there isn't a sufficient statistic you can accumulate for MAD (either version of MAD) that takes less space than the original data.
On the other hand, you also need to use 2-pass algorithms to compute Mean Absolute Deviation, whereas STD can be easily calculated in one pass. And you still need standard deviation as it relates directly to the second moment about the mean.
Also, annoyingly, Median Absolute Deviation competes for the MAD name and is more robust against outliers.
They note that noisy environments don't really help.
I was also a backer.
I think the most compelling example of how bad the controls are is to compare the 'pinball arcade' game they have to the PC or mac version.
On the PC or Mac the flippers work instantaneously and the game is quite fun.
On the Ouya it is unplayable, with half second latencies, it is almost impossible to pull back the plunger to start the ball, etc.
Also, on github you can still tag versions and it'll still put up tar balls you just can't upload some zip file or your own and get it in the same place as the tagged versions.
At a convention the people aren't usually sitting at their computers the whole time huddled over problems they are working on. A hackathon for me is a group setting of people working on code. A convention has them in breakout sessions talking about it, but not writing it.
I'm going to disagree with most of the replies I've seen here so far about just piling on constraints and limitations.
When I go to a hackathon, I am looking for an open forum with interesting people to talk to and people who have their own problems to solve. I get sucked into new problems for 2-3 days and I emerge on the other side with insights into areas I wouldn't have thought of working on before.
I'm not looking for structure from the organizer about what to work on. I and most of the people I know already have a ton of projects in the wings. I'm looking for a good collaborative space to talk to people, people who've brought interesting projects to help, and whiteboard/blackboard space to use for explaining things.
The Haskell hackathons (Hac-Phi and Hac-Boston in particular) have generally followed this format and I love them.
I've gone to other events where someone is trying to harness a hackathon to achieve some particular end and pass out prizes or something, and in general I've been bored out of my mind. If I want to go work with some fixed group of people on some fixed task I can do that. It is called a job.
I'm at a hackathon to generally improve the state of things that the people around me are passionate about and to be exposed to new things.
As a CIO, I viewed my job to be the opposite of everything in this article.
Of course it is good to listen. It is good to be able to interact with anyone on their level of technical expertise and understanding. This advice holds at every level of an organization.
It is also occasionally good to be capable of being demonstrably the most technically competent person in the room. Effective organizations do need the person who can actually ensure there exists an implementble strategy to accomplish the things the CEO is selling the world, and the things the client wants, and who can articulate to vendors exactly why their magic bullet isn't quite what you need. And in many ways as a CIO, your role is to be the one person at that level of management who really understands the ins and outs of how the technology works, how things can improve and how you can adapt to meet the challenges of the organization as a whole.
Sometimes that means being the voice of reason as the curmudgeonly technology guy, but more often it means trying to steer management towards implementable solutions and being able to suggest things that give the other CXO types options they didn't know existed.
Whether facing inward within the organization or outward to clientele or vendors, you need to be able to communicate effectively. One thing this article omits is that when facing outward, it is often good to know when to overload the vendor to get to someone who is more competent to address your concerns, and somewhat more judiciously to be able to out-tech a client's technical guys as well.
Sometimes it _does_ pay to be the smartest person in the room.
They could probably do even better by getting the phone to emit a loud clapping sound, approximating a dirac delta so they could measure the impulse response of the room. The fourier transform of that should have some nice distinctive shapes. On the other hand, that wouldn't be nearly as unobtrusive, and most smart phones have crappy speakers, so you wouldn't get much response.
I used CLEPs for 30 credits of a double B.S. including almost all of my general education requirements. The key with getting most universities to accept CLEPs -- which are mostly targeted at people coming out of highschool with some college level experience -- is to make sure that you haven't taken an official college level course in the department that you plan to CLEP in. I CLEP'd out of literature, american history, sociology, american government, composition, economics ... Basically everything I could. Take AP exams for everything you can as well. I still wound up having to take a couple of classes outside of math and computer science in order to fulfill art and cross-cultural coursework requirements -- the one that stuck out most was african american theatre -- but overall I can't say I'm any worse off for having given the other departments their pound of flesh.
I did have to study for the CLEPs I took to ensure I passed them all, but I was much happier studying sociology for a weekend than for a semester.
The last time after I found myself seriously entangled with a girl, she couldn't stand to be in the same state with me.
*ducks*
I stand corrected. I had at first thought general relativity was being used. I should have RTFA.
That said, while string theory can unify them on paper, string theory comes off as more of a religion than a science given that most of its flavors are completely non-falsifiable, because they have so many degrees of freedom that they can excuse pretty much anything.
It is remarkable in the fact that all of the previous attempts to mix Quantum-"anything" with Relativity have pretty much spectacularly failed.
I'm quite impressed.
Ah then with 'contradistinctions' and 'transubstantiation' he manages to quite thoroughly trounce qwerty. ;) I didn't catch the one hand bit.
What are you calling more efficient?
A left handed dvorak user maxes out at 6 characters with something like 'papaya' or 'yuppie.' ;)
Even a right handed qwertyist at least gets to 9 with 'polyphony.'
By contrast a right handed Dvorak user can't even put the . on the abbreviation for boulevard 'blvd.' Thats what you get for putting all of your eggs (er vowels) in one basket (er hand) and sliding your punctuation around. ;)
We usually call this notion the Curry Howard correspondence.
http://en.wikipedia.org/wiki/Curry-Howard_correspondence
It is an idea used a lot by programmers in languages like Haskell.
In other news, Blizzard is looking forward to record profits in the coming quarters from windfall lawsuits against the government willfully violating their terms of service.
In other other news, after reaching an agreement with the US government, the Chinese government seeks to reach similar terms to investigate Turing-complete gold farmers pointing out that gold farming is an intrinsically human activity and that a bot that could accomplish the same ends would be a great boon to society.
Asked what he was going to do with the award money, the artist said "I'll finally be able to afford Photoshop!" ;)