So, at my most wasteful "always upsize setting", that's a whopping
<$20 for one whole year (looking at their average
spending per customer, I'm probably a big spender!). Can anyone
beat that? That was before. Just now, I check their news pages, and
I would like to take this opportunity to shill for my shared
hosting provider, as a satisfied customer (I have not other relation
with them). I was going to say that these guys offer $1 per G
bandwidth, and $1 per Meg per month disk space, strictly on a
pay-as-you-go basis. What was my hosting fees over the last year? I
don't know, probably around $10 -> that's $7 for the domain name, and
$3 for MySql, leaving a few cents for storage & bandwidth. I'm a very
light user, and well, I'm working on getting more traffic/revenue etc,
but it's a while yet. Ok, now I also have "unlimited account" email
forwarding at $0.02 per day ie anything, ***@mysite.com gets
forwarded.
So, at my most wasteful "always upsize setting", that's a whopping
average
spending per customer, I'm probably a big spender!). Can anyone
beat that? That was before. Just now, I check their news pages, and
found that the bandwidth costs have just been decreased (and check
this out for those rational-minded ones out there) on a log scale!
These discounts aren't monthly or anything; the more you transfer,
the cheaper your service gets, no matter how long it takes. It's also
permanent for the life of your bandwidth account...
This is absolutely not designed to compete with the
bajillion-gigs-for-$9.95-a-month* plans out there. Those plans are
based on overselling, not actual cost of services. We'll never go down
that road. Instead, we've been waiting a long time to get the
purchasing power and volume discounts needed to make hosting even more
affordable for our members, and it's very exciting that after six
years, we're finally here. Languages
supported
(Yes! Haskell, Ruby, Lisp, everything under the sun, as long as it's
not a persistent server model (ie must be "cgi") -- you'll probably
need a VPS for that, or dedicated hosting). Oh, there's ssh too.
> I'm more interested in the possibility that some species of dinosaur > became sentient, built a technological civilization, and then erased > all traces of themselves from the planet (causing mass extinctions in > the process) before moving out into space. It's no more likely than > ape-humping pyramid-building aliens, but sentient space dinosaurs > would be a lot cooler.
In Dr Who, "The Silurians" and "Monsters of the Deep", have this similar plot line.
"In the 1970's we shot a lot of film of this on an air hocky table and took measurements from the photographs to calculate displacement of the objects photographed under a strobe light."
And technology can add to that today. While we have to use technology with our eyes wide open, there's no reason why we cannot be exponentially smarter because of Moore's law. Just as an example, Alan Kay said something like "a different context is worth 80 IQ points" when he talked about very young children intuitively doing calculus (in a quantitative manner too) using the Squeak system. There's video demos and such on the Squeakland website. Additionally, I've found the chatter on the mailing list to be particularly interesting.
Yes, but I had a lot of trouble trying to understand what it was all about (confusing syntax/semantics, at least for me). Nowadays, I have to work on windows too, so it's a mongrel mix of Common Lisp (as much as possible), Emacs, cygwin bash (or bash through ext:run-command in clisp). The problem is that most major applications don't really want to talk to other programs, so this is still quite limited. So it's still not quite the same as my wish, which is a full, live system all the way down to the metal, whatever it may be -- Lisp, Smalltalk, Forth, etc. It's great being able to physically see objects and manipulate them, but at the lowest level, we are still stuck with an opaque OS.
Your questions concern things we can add to today's software to
improve them. Let me flip that around. Why not REMOVE some software to
improve the computing experience for humanity as a whole?
Why, in the 21st century, are we still stuck on 1970's paradigms? Why
Linux? Why Unix? Why do we still work on layered and opaque systems?
Is the dominance of the C-language, Unix-operating-system duo today a
historical accident? Or was it purely evolution: survival of the
fittest (fittest may even be far from the best technological
solution).
Let me play the devil's advocate here by suggesting that we need to
now rethink the old technology that is a systems environment that is
not well integrated and extendable. Operating systems today are 'dead'
and static (once compiled) monolithic programs. To bring dynamism to
the platform, we then layer on abstractions like shells, (bash, csh,
zsh, etc), scripting languages (sed, grep, awk, perl,+ newer friends),
windowing systems, etc. Our compiler toolchain consist of separate
tools that don't talk to each other except through what they see in
front of them, on the 'bit conveyor belt'.
Eg the autotools -- M4, autoconf, automake, configure, make (roughly
in that sequence). Here's a quick test, is there a way we can define a
variable that is checked in each of those stages, to control certain
(compilation, configuration, other settings) options? No. Instead, we
pass on values through multiple languages/interfaces or filesystems in
the form of strings, macros, environment variables, files, etc. Who's
gonna maintain all these software, to fix and sometimes perpetuate(!)
bugs?
May I suggest that this is not a problem with the userspace apps, but
rather a systemic defect in the system we use today. I wish the
operating system is open all the way (and I don't think this conflicts
with the need for security in the system). I wish for a system that is
extensible at runtime, one with a single language all the way, simple
models of abstraction and no unnecessary barriers between programmer
and user. I wish the computer would take care of the details, and help
me do my work. I wish life was good.
Disclaimer: I do not have all the solutions to my complains, so I'm
just a sufferer ranting. Ok, this is also not quite what you asked for,
but it's probably not that offtopic, I'm just off on a slight tangent.
You can do the Fourier transform along each axis in sequence. First, transform along x: f(x,0),f(x,1),f(x,2)... can be transformed separately into f'(u,0),f'(u,1),f'(u,2)... Then transform along y, also separately: f'(0,y),f'(1,y),f'(2,y)... becomes F(0,v),F(1,v),F(2,v)... ie F(u,v).
Returning to the 1D FFT, which is a DFT repeatedly decomposed into smaller DFT's, I don't think it is as parallelisable because although an N-point DFT can be split into 2 N/2-point DFT's, the "communication costs" involved in doing so, and in then combining the output from the 2 DFT's back into one, is prohibitive. Hmmm, will need to think about this... maybe we can return to the original matrix multiplication ie plain DFT (no FFT speedup), which is parallelisable.
What a coincidence, this exact thing (if I have understood you correctly) did happen! The result? The program had to be released under GPL. Read all about it here.
(this seems like a rapidly changing URL - it's basically a file in the clisp project).
Yeah, most of the stuff I recently got interested in follow the same 'framework' - given a set of data (doesn't matter what), you extract some features you are interested in and classify them. The features can have discrete values (sensor-A triggered, item B detected, test C positive), or continuous (humidity D = 90%, length E = 3.4 m, etc). Pass those feature vectors through a blackbox classifier, and attempt to 'fit' the features into a suitable class.
The classes can be discrete (if binary, this could be a detection system: -1 means object not detected, 1 mean object detected, or test passed/positive etc, multivalued would be as explained in this example) in which case we have classification. If the classes were continuous, we have regression (given these features (age, weight, country), the most likely estimate for a person's height is 1.75 m). When you design/train a classifier, you are basically looking for a function f(features) -> class that best describes your observed training data.
So you see, it's all the same to me:). What is hard about images is in how smart you are in extracting those features. All you get is a bunch of pixels after all. The values they take on varies as lighting level, distance and colour varies. Additionally, objects in images can be rotated, reflected/mirrored, scaled, sheared, or distorted by the camera lens & position. Because of this, it's really difficult to classify pixels. It gets worse if all you are getting (as in this application) is low resolution highly compressed images (because of the additional artifacts).
The Markov text generator you mentioned could be generalised to techniques dealing with N-grams. If you see for example Shannon's landmark paper about entropy - he talked about the very same technique for generating plausible looking paragraphs of text, so the Mark V. Shaney (ref wikipedia) algorithm has probably been around in some form or other since 'antiquity'.
--- RANT ---
As for Bayesian logic, yeah it's all cool, but there really is a HUGE class of techniques and algorithms one needs to draw upon to do AI. It takes forever to learn them all, but you almost need to know them all to be able to confidently claim that any technique is optimal for a particular situation... add to this the sometime fraudulent claims, myths and hype that has been circulating since the AI winter days, and you really have be careful. I'm a stauch Bayesian, and for the life of me could not get fuzzy logic. Why? There's the Cox axioms for reasoning with uncertainty, that forms the foundation for probability, something like probabilities must be greater than 0, must sum for mutually exclusive things, and I think there's probably something for products independent events - look it up. Thing is, there's a thing called the Dutch book that guarantees a loss to anyone who violates the Cox axioms in their reasoning. IE if you are playing the financial markets and you take two probabilities/beliefs/certainty factors/membership functions and use their min/max for and/or instead of products and sums (caveats apply), then doesn't this violate the Cox axiom? Doesn't that mean someone else using Bayesian inference would eat your lunch? Those fuzzy logicists always sidestep this by saying a fuzzy membership function is not a belief or probability. Well, it looks like one, it quacks like one, and it doesn't satisfy Cox's axioms. What gives?
--- END OFFTOPIC RANT/CHALLENGE --- PS I thinks they're called Cox's axioms -> I am not a mathematician.
Warren Buffet claims he doesn't know jack about technology (and that may well be true), but he knows businesses, and he knows his business well (he's basically an Encyclopaedia Capitalisma). Read from his lecture at Notre Dame. No one else seems to be saying this, but technological or scientific advancement is bad for preserving capital (we not talking millions, we're talking billions). For us small fry trying to eke out somewhere in the millions, technology probably isn't too destructive to capital, but hey, what do I know.
Agony vs. Ecstasy Businesses:
Example 1
It does make a difference what kind of a business you get associated
with. For that reason I've set forth in this little handout Company A
and Company E. I'm not going to tell you for the moment what these
companies are. I'm going to tell you one thing about the two
companies. One of the companies, to the point of where this cuts off,
lost its investors more money than virtually any business in the
world. The other company made its owner more money than virtually any
company in the world. So one of these two companies, Company A and
Company E, has made one of its owners one of the five wealthiest
people in the world, while the other company made its owners
appreciably poorer, probably more so than any other company to that
point in time.
Now I'll tell you a little bit about these companies (we're leading up
to the question of whether the business makes a difference). Company A
had thousands of MBAs working for it. Company E had none. I wanted to
get your attention. Company A had all kinds of employee benefit
programs, stock options, pensions, the works. Company E never had
stock options. Company A had thousands of patents - they probably held
more patents than just about any company in the United States. Company
E never invented anything. Company A's product improved dramatically
in this period, Company E's product just sat.
So far, based on what I've told you, does anybody have any idea of
which company was the great success, and why?
If you get to buy one of these two companies, and this is all you
know, and you get to ask me one question to decide on which one to
buy. If you ask me the right question, you will probably make the
right decision about the company's stock, and one will make you
enormously wealthy.
[Audience asks questions]
Both companies make products used every day. They started as
necessities, highly useful, nothing esoteric about either one,
although company A does have all these patents. There's more
technology involved in company A.
[How many companies compete with either one?]
Good question, very good question. In effect, neither company had any
competition. And that might differentiate in some cases. Well, I'll
tell you a little more about it. Company A is known as company A
because it was in agony, and Company E, as Company E, because it was
in ecstasy. Company A is American Telephone and Telegraph. I've
omitted eight zeros on the left hand side, and the American Telephone
and Telegraph Company, at the end of 1979, was selling for $10 billion
less than the shareholders had either put in or left in the
business. In other words, if shareholder's equity was "X" the market
value was X minus $10 billion. So the money that shareholders had put
in, or left in, the business had shrunk by $10 billion in terms of
market value.
Company E, the excellent company, I left off only six zeros. And that
happens to be a company called Thompson Newspapers. Thomson
Newspapers, which most of you have probably never heard of, actually
owns about 5% of the newspapers in the United States. But they're all
small ones. And, as I said, it has no MBAs, no stock options - still
doesn't - and it made its owner, Lord Thompson. He wasn't Lord
Thompson when he started - he started with 1,500 bucks in North Bay,
Ontario buying a little radio station but, when he got to be one of
the five riche
This is not a beefed up server - it's a very old computer, and is used only for development. Remember that there are 500 simultaneous connections, with the test itself also running on the same machine, and I'm also running a shitload of other stuff (X, mozilla-firefox - heaps of tabs open, Lisp (this one shows hundreds of megs of memory usage, but I've read that for this kind of applications, top is a deceptive measure), emacs etc).
The benchmarks I've shown are from running "ab". I was going to show all program outputs, had to edit it extensively (and finally gave up because I got tired of negotiating with the lameness filter) but it seems accurate. Do you think it could be much faster? Here're some more info:
cpuinfo:
processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 8
model name : AMD Athlon(tm) XP 1700+
stepping : 1
cpu MHz : 1466.795
cache size : 256 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow
bogomips : 2916.35
uname -a gives Linux leia 2.6.12-9-386 #1 Mon Oct 10 13:14:36 BST 2005 i686 GNU/Linux
And finally, it's relative performance that counts, and for the other benchmark I show, where the outperformance is in the orders of magnitude, I wouldn't worry about processor speed/software versions or other configurations.
Oh wait, do you mean something else? Here, it's 500 concurrent requests, so at any time, there's only 500 connections. After 5000 connections have passed, we get this. Check out the output from ab, or read man ab, for a rough handle on what I have shown before.
I cannot speculate, but I can look at what people are doing today. One thing that I have noticed, is the widespread research into, with compelling arguments, for massively multithreaded programming
techniques. See Erlang for example. It is designed right from the beginning for this sort of problem - high throughput, high reliability, high uptime telephony networks.
That's an order of magnitude increase in "performance" (depends on what you mean by performance". I thought I'll do a casual informal test of my own, with a decent static file size (instead of the 1 byte
used in that benchmark)
Server Software: Yaws/1.56
Document Length: 402 bytes
Concurrency Level: 500
Time taken for tests: 8.480740 seconds
Complete requests: 5000
Requests per second: 589.57 [#/sec] (mean)
Time per request: 848.074 [ms] (mean)
Server Software: Apache/2.0.54
Document Length: 402 bytes
Concurrency Level: 500
Time taken for tests: 29.787216 seconds
Complete requests: 5000
Requests per second: 167.86 [#/sec] (mean)
Time per request: 2978.722 [ms] (mean)
Output edited to get past lameness filter.
Err crap, I could have sworn the first time I tried this, when Yaws was first installed, its performance was worse! Oh well, perhaps it's something I've inadvertently done since then. Could have been due to my computer reboot (this is a desktop PC). It seems I've proven my point, although I was trying to disprove it. Standard caveats regarding benchmarks apply. Both servers are default Ubuntu installs with no configuration changes - I didn't compile anything manually.
Additionally it has also been noted that:
> Linus Torvalds: 100k threads at once is crazy
Using Posix style threads, I'd have to agree. Posix threads were just not designed with this level of usage in mind. Which is why concurrent lanugages like Erlang and Mozart/Oz don't use Posix threads.
Well, that's where it could be headed anyway - a multiprocessor system with green threads (ie simulated threads, like Java ones) implementing massive concurency and redundancy. Some prototypes for systems like
this are already available, and being used.
Cheers.
Err maybe not. I think my ignorance is beginning to show. The problem might be rather - continuations allow you to escape many times, but unwind-protect should only run once. So if you wrap some code with unwind-protect, it is a postlude to that code, and should run when you exit (either a normal exit, or by non-local ones like throwing exceptions) the wrapped code. Continuations inside that wrapped block of code can "continue" many times, so we have no way of knowing when we are done, so there's no straight answer to when we can run the postlude code. Whatever - if anyone's really interested, don't listen to me - do your own reading.
Continuations can be used to build everything (don't know, maybe everything), but from what I've managed to gather, there are problems with continuations as they are used in Scheme at the moment (don't know about any other language).
I've been trying to read through and understand this, but haven't got far yet. Basic idea is I guess "how does a continuation escape from an unescapable unwind-protect?". Unwind-protect is like "finally" in Java, it has to run *always* - but continuations never return, so it's like when an immovable object meets an irresistable force.
Additionally, there's added issues with general error handling and other stuff - how can we do this efficiently and without getting in the way of the programmer - many of these issues are subjective, so what I meant was not the same thing you were thinking I guess.
I don't pretend to grok continuations (that's what Seaside does I hear), but an article I have found to be really illuminating explains this in quasi-Basic. The Basic-ish code in that example by the way, is written in what we call continuation passing scheme (CPS). That's basically the extra function tacked onto the argument list passed to functions [instead of f(arg1,arg2), we call f(arg1,arg2, c) where c is the continuation. f does not return as such, but calls c when it is ready to return. Instead of "return some_value;", we call "c(some_value)"].
Of course, continuations are not the solution to everything, but there's a lesson to learn from their usage.
CPS is often what functional programming languages compile into - no one writes code like that manually. You take a program, transform it into CPS, keep doing that, resubstituting variables names, expanding macros, and simplifying functions as necessary, until you are left with a couple of variables (-> registers), and a whole lot of assembler instructions - it has been compiled into assembly language. Now that's a nice compiler!
This is reminds me of the story about when Amazon went into auctions, taking on ebay. That didn't work out. Why? The customer behaviour is just different - you go to Amazon, you see book, you click. There's suggestions for other books etc, and you might or might not get those as well along the way. You deal with a company with a known reputation. Auctions is about bargain hunting, and assessing the reputation of whoever you are buying from. The consumer behaviour is just different.
Now, how is this grocery store any different? Can they give me a list of statistically improbable items in my shopping list (fertiliser, detergent, ammonium nitrate, charcoal, starch, paraffin oil...:);) just joking), or suggest grocery items? Or provide consumer reviews of products? Can that work? It might, but it sounds hard. There doesn't seem to be a compelling case.
The link I provided is very entertaining by the way. If interested, you want to start here. And read the whole blog, it's full of interesting tidbits (I'm not done with it yet). Greg's cool! He's one of us. (But wait there's more: here's a similar one for Google too! Doug's a marketing guy, and you can read Ron Garret's posts for the technical stuff.)
That was all typed into the REPL, ie interactively. The function was incrementally compiled, so read time, compile time, and runtime, while distinct, are all available at the same time. EDX and EDI contains the first and second arguments to a function, while EAX contains the funobj, and ECX the argument count.
I know you said no IDE's, but if you merge well with Emacs, it can do the work too. Emacs is not exactly a heavy weight, depending on how you install it (it's often built into your distro anyway). It uses this thing (I don't really know it that well to be honest) called tags - ctags or etags.
Basically you run etags (check your man pages) from the command line that will parse through your source files and create a lookup table in a file (name TAGS by default I think). While browsing the source file, you just have to position the cursor at the right symbol, and press M-. (that's usually Alt-DOT) and it'll take you to the function definition. (vim has a similar thing)
I haven't used it too much, since I'm a Lisper, and the Slime development environment (Emacs addon mode) for Lisp has a similar thing (and it doesn't need to create any tables beforehand) that also provides a "stack-like" functionality. That means you can jump to the function definition, then pop back to where you were. This can be handy for quick detours just to lookup small functions for example.
The advantage here is that you have the all the files locally, so it's faster than browsing through a web interface (html-ized source files, like the Sourceforge CVS frontend - I still use that a lot, and it is SLOW), and you can also edit the source (just a bonus).
It's usually better than arriving late to the party with a perfect product (ironically, if you're *really* good, this can be the deciding factor for something like Lisp).
I'm curious. Can you elaborate? I cannot really understand what you mean here. If you're *really* good, should you or shouldn't you use Lisp, and why?
It wasn't that long ago (err, wow, 10 years, maybe that's long) that the first extrasolar planet was discovered. I still remember that news announcement I watched on TV...
Anyway, since the discovery of those 3 planets, another planet has been found. Check out the exoplanet encyclopedia (my favourite exoplanets site). It has a catalog with all the data of those planets, some with uncertainty factors. Discovery method, size, catalogue number, the whole lot. Try chucking all that into a spread-sheet, and plot some scatter graphs. Should be a lotta fun. The last time I tried this, it was a bit problematic because the masses are not really known (for planets discovered using spectral shifts), but are merely minimum (maximum?) limits only. But still, an order of magnitude plot could be fun.
Anyway, the 3 planets are already in the catalogue under HD 69830. Don't forget to check out this one as well. Exciting times. I look forward to 200 planets!
I don't know too much about chimp sexual habits, but we humans sure are a kinky bunch to boot.
No we're not, not even close. There are several (2? 3?) species of chimps, and they have distinct sexual behaviours. The most promiscuous chimp species is the Bonobo. In the bonobo, copulation is extremely common, and form the backbone of their social bonding fabric. See wikipedia, or read the Third Chimpanzee by Jared Diamond for example. Suffice to say, we humans are uptight conservative puritans by comparison.
Slashdot Mirror
Thanks for the reference.
So, at my most wasteful "always upsize setting", that's a whopping <$20 for one whole year (looking at their average spending per customer, I'm probably a big spender!). Can anyone beat that? That was before. Just now, I check their news pages, and
I would like to take this opportunity to shill for my shared hosting provider, as a satisfied customer (I have not other relation with them). I was going to say that these guys offer $1 per G bandwidth, and $1 per Meg per month disk space, strictly on a pay-as-you-go basis. What was my hosting fees over the last year? I don't know, probably around $10 -> that's $7 for the domain name, and $3 for MySql, leaving a few cents for storage & bandwidth. I'm a very light user, and well, I'm working on getting more traffic/revenue etc, but it's a while yet. Ok, now I also have "unlimited account" email forwarding at $0.02 per day ie anything, ***@mysite.com gets forwarded.
So, at my most wasteful "always upsize setting", that's a whopping average spending per customer, I'm probably a big spender!). Can anyone beat that? That was before. Just now, I check their news pages, and found that the bandwidth costs have just been decreased (and check this out for those rational-minded ones out there) on a log scale!
I quote
These discounts aren't monthly or anything; the more you transfer, the cheaper your service gets, no matter how long it takes. It's also permanent for the life of your bandwidth accountYou're thinking Friedrich Kekule and the benzene ring?
> I'm more interested in the possibility that some species of dinosaur
> became sentient, built a technological civilization, and then erased
> all traces of themselves from the planet (causing mass extinctions in
> the process) before moving out into space. It's no more likely than
> ape-humping pyramid-building aliens, but sentient space dinosaurs
> would be a lot cooler.
In Dr Who, "The Silurians" and "Monsters of the Deep", have this
similar plot line.
"In the 1970's we shot a lot of film of this on an air hocky table and took measurements from the photographs to calculate displacement of the objects photographed under a strobe light."
And technology can add to that today. While we have to use technology with our eyes wide open, there's no reason why we cannot be exponentially smarter because of Moore's law. Just as an example, Alan Kay said something like "a different context is worth 80 IQ points" when he talked about very young children intuitively doing calculus (in a quantitative manner too) using the Squeak system. There's video demos and such on the Squeakland website. Additionally, I've found the chatter on the mailing list to be particularly interesting.
Also see this powerpoint presentation by Abe Lincoln in 1863 :).
Yes, but I had a lot of trouble trying to understand what it was all about (confusing syntax/semantics, at least for me). Nowadays, I have to work on windows too, so it's a mongrel mix of Common Lisp (as much as possible), Emacs, cygwin bash (or bash through ext:run-command in clisp). The problem is that most major applications don't really want to talk to other programs, so this is still quite limited. So it's still not quite the same as my wish, which is a full, live system all the way down to the metal, whatever it may be -- Lisp, Smalltalk, Forth, etc. It's great being able to physically see objects and manipulate them, but at the lowest level, we are still stuck with an opaque OS.
Sup Eli. I remember you from comp.lang.lisp!
:)
Yes, I am (half) aware of the practicalities of todays world. However, as a Lisper, I find your lack of faith disturbin'.
Your questions concern things we can add to today's software to improve them. Let me flip that around. Why not REMOVE some software to improve the computing experience for humanity as a whole?
Why, in the 21st century, are we still stuck on 1970's paradigms? Why Linux? Why Unix? Why do we still work on layered and opaque systems? Is the dominance of the C-language, Unix-operating-system duo today a historical accident? Or was it purely evolution: survival of the fittest (fittest may even be far from the best technological solution).
Let me play the devil's advocate here by suggesting that we need to now rethink the old technology that is a systems environment that is not well integrated and extendable. Operating systems today are 'dead' and static (once compiled) monolithic programs. To bring dynamism to the platform, we then layer on abstractions like shells, (bash, csh, zsh, etc), scripting languages (sed, grep, awk, perl,+ newer friends), windowing systems, etc. Our compiler toolchain consist of separate tools that don't talk to each other except through what they see in front of them, on the 'bit conveyor belt'.
Eg the autotools -- M4, autoconf, automake, configure, make (roughly in that sequence). Here's a quick test, is there a way we can define a variable that is checked in each of those stages, to control certain (compilation, configuration, other settings) options? No. Instead, we pass on values through multiple languages/interfaces or filesystems in the form of strings, macros, environment variables, files, etc. Who's gonna maintain all these software, to fix and sometimes perpetuate(!) bugs?
May I suggest that this is not a problem with the userspace apps, but rather a systemic defect in the system we use today. I wish the operating system is open all the way (and I don't think this conflicts with the need for security in the system). I wish for a system that is extensible at runtime, one with a single language all the way, simple models of abstraction and no unnecessary barriers between programmer and user. I wish the computer would take care of the details, and help me do my work. I wish life was good.
Disclaimer: I do not have all the solutions to my complains, so I'm just a sufferer ranting. Ok, this is also not quite what you asked for, but it's probably not that offtopic, I'm just off on a slight tangent.
Yes, that's what I thought too, at first, but the original poster was
actually referring to N dimensional FFT's, and they are separable.
Eg, 2D DFT:
\int_{-\infty}^{\infty} \int_{-\infty}^{\infty} f(x,y) e^{-i 2 \pi (ux+vy)} dx\ dy
= \int_{-\infty}^{\infty} \int_{-\infty}^{\infty} f(x,y) e^{-i 2 \pi ux) dx e^{-i 2 \pi vy} dy
You can do the Fourier transform along each axis in sequence. First,
transform along x: f(x,0),f(x,1),f(x,2)... can be transformed
separately into f'(u,0),f'(u,1),f'(u,2)... Then transform along y,
also separately: f'(0,y),f'(1,y),f'(2,y)... becomes
F(0,v),F(1,v),F(2,v)... ie F(u,v).
Returning to the 1D FFT, which is a DFT repeatedly decomposed into
smaller DFT's, I don't think it is as parallelisable because although
an N-point DFT can be split into 2 N/2-point DFT's, the "communication
costs" involved in doing so, and in then combining the output from the
2 DFT's back into one, is prohibitive. Hmmm, will need to think about
this... maybe we can return to the original matrix multiplication ie
plain DFT (no FFT speedup), which is parallelisable.
What a coincidence, this exact thing (if I have understood you correctly) did happen! The result? The program had to be released under GPL. Read all about it here. (this seems like a rapidly changing URL - it's basically a file in the clisp project).
Yeah, most of the stuff I recently got interested in follow the same
:). What is hard about images is
'framework' - given a set of data (doesn't matter what), you extract
some features you are interested in and classify them. The features
can have discrete values (sensor-A triggered, item B detected, test C
positive), or continuous (humidity D = 90%, length E = 3.4 m,
etc). Pass those feature vectors through a blackbox classifier, and
attempt to 'fit' the features into a suitable class.
The classes can be discrete (if binary, this could be a detection
system: -1 means object not detected, 1 mean object detected, or test
passed/positive etc, multivalued would be as explained in this
example) in which case we have classification. If the classes were
continuous, we have regression (given these features (age, weight,
country), the most likely estimate for a person's height is 1.75
m). When you design/train a classifier, you are basically looking for
a function f(features) -> class that best describes your observed
training data.
So you see, it's all the same to me
in how smart you are in extracting those features. All you get is a
bunch of pixels after all. The values they take on varies as lighting
level, distance and colour varies. Additionally, objects in images can
be rotated, reflected/mirrored, scaled, sheared, or distorted by the
camera lens & position. Because of this, it's really difficult to
classify pixels. It gets worse if all you are getting (as in this
application) is low resolution highly compressed images (because of
the additional artifacts).
The Markov text generator you mentioned could be generalised to
techniques dealing with N-grams. If you see for example Shannon's
landmark paper about entropy - he talked about the very same technique
for generating plausible looking paragraphs of text, so the Mark
V. Shaney (ref wikipedia) algorithm has probably been around in some
form or other since 'antiquity'.
--- RANT ---
As for Bayesian logic, yeah it's all cool, but there really is a HUGE
class of techniques and algorithms one needs to draw upon to do AI. It
takes forever to learn them all, but you almost need to know them all
to be able to confidently claim that any technique is optimal for a
particular situation... add to this the sometime fraudulent claims,
myths and hype that has been circulating since the AI winter days, and
you really have be careful. I'm a stauch Bayesian, and for the life of
me could not get fuzzy logic. Why? There's the Cox axioms for
reasoning with uncertainty, that forms the foundation for probability,
something like probabilities must be greater than 0, must sum for
mutually exclusive things, and I think there's probably something for
products independent events - look it up. Thing is, there's a
thing called the Dutch book that guarantees a loss to anyone who
violates the Cox axioms in their reasoning. IE if you are playing the
financial markets and you take two probabilities/beliefs/certainty
factors/membership functions and use their min/max for and/or instead
of products and sums (caveats apply), then doesn't this violate the
Cox axiom? Doesn't that mean someone else using Bayesian inference
would eat your lunch? Those fuzzy logicists always sidestep this by
saying a fuzzy membership function is not a belief or
probability. Well, it looks like one, it quacks like one, and it
doesn't satisfy Cox's axioms. What gives?
--- END OFFTOPIC RANT/CHALLENGE ---
PS I thinks they're called Cox's axioms -> I am not a mathematician.
Warren Buffet claims he doesn't know jack about technology (and that may well be true), but he knows businesses, and he knows his business well (he's basically an Encyclopaedia Capitalisma). Read from his lecture at Notre Dame. No one else seems to be saying this, but technological or scientific advancement is bad for preserving capital (we not talking millions, we're talking billions). For us small fry trying to eke out somewhere in the millions, technology probably isn't too destructive to capital, but hey, what do I know.
This is not a beefed up server - it's a very old computer, and is used only for development. Remember that there are 500 simultaneous connections, with the test itself also running on the same machine, and I'm also running a shitload of other stuff (X, mozilla-firefox - heaps of tabs open, Lisp (this one shows hundreds of megs of memory usage, but I've read that for this kind of applications, top is a deceptive measure), emacs etc).
The benchmarks I've shown are from running "ab". I was going to show all program outputs, had to edit it extensively (and finally gave up because I got tired of negotiating with the lameness filter) but it seems accurate. Do you think it could be much faster? Here're some more info:
cpuinfo:
processor : 0
vendor_id : AuthenticAMD
cpu family : 6
model : 8
model name : AMD Athlon(tm) XP 1700+
stepping : 1
cpu MHz : 1466.795
cache size : 256 KB
fdiv_bug : no
hlt_bug : no
f00f_bug : no
coma_bug : no
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mmxext 3dnowext 3dnow
bogomips : 2916.35
uname -a gives Linux leia 2.6.12-9-386 #1 Mon Oct 10 13:14:36 BST 2005 i686 GNU/Linux
And finally, it's relative performance that counts, and for the other benchmark I show, where the outperformance is in the orders of magnitude, I wouldn't worry about processor speed/software versions or other configurations.
Oh wait, do you mean something else? Here, it's 500 concurrent requests, so at any time, there's only 500 connections. After 5000 connections have passed, we get this. Check out the output from ab, or read man ab, for a rough handle on what I have shown before.
I cannot speculate, but I can look at what people are doing today. One thing that I have noticed, is the widespread research into, with compelling arguments, for massively multithreaded programming techniques. See Erlang for example. It is designed right from the beginning for this sort of problem - high throughput, high reliability, high uptime telephony networks.
As a rough benchmark, someone's got this.
That's an order of magnitude increase in "performance" (depends on what you mean by performance". I thought I'll do a casual informal test of my own, with a decent static file size (instead of the 1 byte used in that benchmark)
Server Software: Yaws/1.56
Document Length: 402 bytes
Concurrency Level: 500
Time taken for tests: 8.480740 seconds
Complete requests: 5000
Requests per second: 589.57 [#/sec] (mean)
Time per request: 848.074 [ms] (mean)
Server Software: Apache/2.0.54
Document Length: 402 bytes
Concurrency Level: 500
Time taken for tests: 29.787216 seconds
Complete requests: 5000
Requests per second: 167.86 [#/sec] (mean)
Time per request: 2978.722 [ms] (mean)
Output edited to get past lameness filter.
Err crap, I could have sworn the first time I tried this, when Yaws was first installed, its performance was worse! Oh well, perhaps it's something I've inadvertently done since then. Could have been due to my computer reboot (this is a desktop PC). It seems I've proven my point, although I was trying to disprove it. Standard caveats regarding benchmarks apply. Both servers are default Ubuntu installs with no configuration changes - I didn't compile anything manually.
Additionally it has also been noted that:
Well, that's where it could be headed anyway - a multiprocessor system with green threads (ie simulated threads, like Java ones) implementing massive concurency and redundancy. Some prototypes for systems like this are already available, and being used. Cheers.
Err maybe not. I think my ignorance is beginning to show. The problem might be rather - continuations allow you to escape many times, but unwind-protect should only run once. So if you wrap some code with unwind-protect, it is a postlude to that code, and should run when you exit (either a normal exit, or by non-local ones like throwing exceptions) the wrapped code. Continuations inside that wrapped block of code can "continue" many times, so we have no way of knowing when we are done, so there's no straight answer to when we can run the postlude code. Whatever - if anyone's really interested, don't listen to me - do your own reading.
Continuations can be used to build everything (don't know, maybe everything), but from what I've managed to gather, there are problems with continuations as they are used in Scheme at the moment (don't know about any other language).
I've been trying to read through and understand this, but haven't got far yet. Basic idea is I guess "how does a continuation escape from an unescapable unwind-protect?". Unwind-protect is like "finally" in Java, it has to run *always* - but continuations never return, so it's like when an immovable object meets an irresistable force.
Additionally, there's added issues with general error handling and other stuff - how can we do this efficiently and without getting in the way of the programmer - many of these issues are subjective, so what I meant was not the same thing you were thinking I guess.
I don't pretend to grok continuations (that's what Seaside does I hear), but an article I have found to be really illuminating explains this in quasi-Basic. The Basic-ish code in that example by the way, is written in what we call continuation passing scheme (CPS). That's basically the extra function tacked onto the argument list passed to functions [instead of f(arg1,arg2), we call f(arg1,arg2, c) where c is the continuation. f does not return as such, but calls c when it is ready to return. Instead of "return some_value;", we call "c(some_value)"].
Of course, continuations are not the solution to everything, but there's a lesson to learn from their usage.
CPS is often what functional programming languages compile into - no one writes code like that manually. You take a program, transform it into CPS, keep doing that, resubstituting variables names, expanding macros, and simplifying functions as necessary, until you are left with a couple of variables (-> registers), and a whole lot of assembler instructions - it has been compiled into assembly language. Now that's a nice compiler!
This is reminds me of the story about when Amazon went into auctions, taking on ebay. That didn't work out. Why? The customer behaviour is just different - you go to Amazon, you see book, you click. There's suggestions for other books etc, and you might or might not get those as well along the way. You deal with a company with a known reputation. Auctions is about bargain hunting, and assessing the reputation of whoever you are buying from. The consumer behaviour is just different.
Now, how is this grocery store any different? Can they give me a list of statistically improbable items in my shopping list (fertiliser, detergent, ammonium nitrate, charcoal, starch, paraffin oil... :) ;) just joking), or suggest grocery items? Or provide consumer reviews of products? Can that work? It might, but it sounds hard. There doesn't seem to be a compelling case.
The link I provided is very entertaining by the way. If interested, you want to start here. And read the whole blog, it's full of interesting tidbits (I'm not done with it yet). Greg's cool! He's one of us. (But wait there's more: here's a similar one for Google too! Doug's a marketing guy, and you can read Ron Garret's posts for the technical stuff.)
Whack!! Language implementation and language design are two separate
things.
CL-USER> (lisp-implementation-type)
"SBCL"
CL-USER> (lisp-implementation-version)
"0.9.9"
CL-USER> (defun test (x y)
(declare (type fixnum x y)
(optimize (speed 3) (debug 0) (safety 0)))
(the fixnum (+ x y)))
STYLE-WARNING: redefining TEST in DEFUN
TEST
CL-USER: (disassemble #'test)
---Slightly modified for readability----------
ADD EDX, EDI ; no-arg-parsing entry point
MOV ECX, [EBP-8]
MOV EAX, [EBP-4]
ADD ECX, 2
MOV ESP, EBP
MOV EBP, EAX
JMP ECX
NOP
BREAK 10 ; error trap
BYTE #X02
BYTE #X18 ; INVALID-ARG-COUNT-ERROR
BYTE #XCD ; EBX
NIL
CL-USER>
That was all typed into the REPL, ie interactively. The function was
incrementally compiled, so read time, compile time, and runtime, while
distinct, are all available at the same time. EDX and EDI contains the
first and second arguments to a function, while EAX contains the
funobj, and ECX the argument count.
Compare the output from gcc -O3 -s
int test(int x, int y){
return x + y;
}
becomes:
test:
pushl %ebp
movl %esp, %ebp
movl 12(%ebp), %eax
addl 8(%ebp), %eax
popl %ebp
ret
At this stage, which output is better becomes a question of whose
compiler is better, and has nothing to do with the fundamental design
of a language.
Other implementations of common lisp might do things differently, for
example, clisp compiles to bytecodes that runs on a virtual machine.
At this moment, Karma Farmer was enlightened. (I hope).
PS: Lameness filter sucks.
I know you said no IDE's, but if you merge well with Emacs, it can do the work too. Emacs is not exactly a heavy weight, depending on how you install it (it's often built into your distro anyway). It uses this thing (I don't really know it that well to be honest) called tags - ctags or etags.
Basically you run etags (check your man pages) from the command line that will parse through your source files and create a lookup table in a file (name TAGS by default I think). While browsing the source file, you just have to position the cursor at the right symbol, and press M-. (that's usually Alt-DOT) and it'll take you to the function definition. (vim has a similar thing)
I haven't used it too much, since I'm a Lisper, and the Slime development environment (Emacs addon mode) for Lisp has a similar thing (and it doesn't need to create any tables beforehand) that also provides a "stack-like" functionality. That means you can jump to the function definition, then pop back to where you were. This can be handy for quick detours just to lookup small functions for example.
The advantage here is that you have the all the files locally, so it's faster than browsing through a web interface (html-ized source files, like the Sourceforge CVS frontend - I still use that a lot, and it is SLOW), and you can also edit the source (just a bonus).
I'm curious. Can you elaborate? I cannot really understand what you mean here. If you're *really* good, should you or shouldn't you use Lisp, and why?
It wasn't that long ago (err, wow, 10 years, maybe that's long) that the first extrasolar planet was discovered. I still remember that news announcement I watched on TV...
Anyway, since the discovery of those 3 planets, another planet has been found. Check out the exoplanet encyclopedia (my favourite exoplanets site). It has a catalog with all the data of those planets, some with uncertainty factors. Discovery method, size, catalogue number, the whole lot. Try chucking all that into a spread-sheet, and plot some scatter graphs. Should be a lotta fun. The last time I tried this, it was a bit problematic because the masses are not really known (for planets discovered using spectral shifts), but are merely minimum (maximum?) limits only. But still, an order of magnitude plot could be fun.
Anyway, the 3 planets are already in the catalogue under HD 69830. Don't forget to check out this one as well. Exciting times. I look forward to 200 planets!
I don't know too much about chimp sexual habits, but we humans sure are a kinky bunch to boot.
No we're not, not even close. There are several (2? 3?) species of chimps, and they have distinct sexual behaviours. The most promiscuous chimp species is the Bonobo. In the bonobo, copulation is extremely common, and form the backbone of their social bonding fabric. See wikipedia, or read the Third Chimpanzee by Jared Diamond for example. Suffice to say, we humans are uptight conservative puritans by comparison.