"Interestingly" relevance for machine learning
on
Google Experiments
·
· Score: 2, Informative
In the spirit of automated classification and machine learning, I tried searching for a set containing "boosting", "SVM" and "bagging" (without "bagging", nothing new is found; another problem).
Results: "SVM", "Bagging", "Boosting", "stacking" and "Other methods".
Clicking on either of the 2 new links ("stacking" or "Other methods") takes me to the normal Google search on the term. This is of course not useful -- I need the terms in the machine learning context, which I cannot get.
Especially in the case of "Other methods", it would be nice to be able to get to the page Google had in mind...
That's implied by universality. Loosely, a CA is universal if it can compute (in some suitable encoding) any computable function (say, any function which can be computed by a Turing machine, or by a suitably unrestricted computer version of Fortran). The big brouhaha comes from the existence of CAs which are indeed universal (and Wolfram showed that many CAs are indeed universal).
Directly from the halting problem, e.g., you can get that it is uncomputable to decide, given a universal CA and an input configuration, whether the configuration "stabilizes" (remains unchanged after some iteration). Rice's theorem says that it is uncomputable to decide, given a universal machine and any input (say, a universal CA and an initial configuration), anything interesting about the output.
Loosely speaking, "the only way to find out the final state of a cellular automaton is to run it".
Read the Inside story quoted by 2600. Closing your eyes is apparently not covered under the current contract; the other thing is a grey area, but evidently broadcasters are currently lax about enforcement:
CW: What if you have to go to the bathroom or get up to get a Coke?
JK: I guess there's a certain amount of tolerance for going to the bathroom. But if you formalize it and you create a device that skips certain second increments, you've got that only for one reason, unless you go to the bathroom for 30 seconds. They've done that just to make it easy for someone to skip a commercial.
I'm sure the FSF people have gone over this a million times (e.g. in gnu.misc.discuss), but here it is, again. The GPL doesn't force you to do anything. It only prohibits you from doing some things. Quoting from the article:
Almost everyone who uses GPL'd software from day to day needs no license, and accepts none. The GPL only obliges you if you distribute software made from GPL'd code, and only needs to be accepted when redistribution occurs. And because no one can ever redistribute without a license, we can safely presume that anyone redistributing GPL'd software intended to accept the GPL. After all, the GPL requires each copy of covered software to include the license text, so everyone is fully informed.
Just like any other software, you cannot redistribute GPL'd software except by certain license provisions. You cannot distribute derivative works of GPL'd software except by certain license provisions.
That is all. When you use Emacs, you aren't forced to do anything. When you sell (or otherwise give) Emacs to someone else, the GPL kicks in. If I download an e-book from somewhere (or buy a CD), I cannot distribute the data as I see fit -- I must follow the license under which I got it. Software is the same. GPL'd software is also no different.
Of course this is untrue. Using flawed Mathematics, one can do anything. Specifically, Gödel showed that a first-order logic axiomatization of any "sufficiently complex" system (arithmetic is more than complex enough) is either inconsistent or incomplete. What you describe is incompleteness. But of course any system which can prove P and also ~P (i.e. is inconsistent) can prove anything (and its negation); you don't need Gödel for that!
As I always say, "if 2+2=5, then the Poincar Conjecture is true".
We know that Einstein's theory disagrees grossly with what we observe. It's not a question of if we can replicate this experimentally but how.
You're saying that there are observations in the real world that grossly disagree with relativity, but that you cannot replicate them experimentally.
Barring what goes on with the very small (quantum effects), what on earth are you talking about? I've never heard of any non-quantum contradictions to relativity. Of course quantum stuff works in the lab; could you please give references to such gross disagreements?
Note that the rapidly-spinning NASA antigravity is hardly claimed to be quantum, so those contradictions don't really help the NASA claimants, either.
That's not a problem, of course. Here's how you code in ARIELSOL: Unless you want to use the "UTM shortcut" (patent pending), just write out your program in ISO standard C. If you decide to take advantage of my NEW! UNIQUE! "UTM shortcut technology", your program should be the single bit `0' (note that this is not a legal C program).
The ARIELSOL interpreter (forthcoming) either compiles and runs your program as though it were a C program (if it's not the single bit 0), OR, if it is the single bit 0, runs itself on the input. Voila! Universality guaranteed, and in the same encoding.
I therefore claim that ARIELSOL encodes its universal machine in a single bit.
(I also claim that there is no Nobel prize in mathematics...)
I still don't get it. If I'm a disgruntled employee (say the company just collapsed and I've just been laid off and feel cheated), what's to stop me making a copy of any email to which I have access?
Saying "secure server" and "secure client" doesn't cut it. As long as I have reasonable access to my computer, I can make a copy. If the computer can display it for me to read, I can copy it.
Surely SeaLand protects against something else completely!
Perl may have much syntactic ugliness, but at least it gets variable binding right.
That's one opinion. Another opinion is that Python supports only local variables, and that global variables are accepted as syntaxic sugar (useful for modules and functions defined out of classes). Some people think that allowing only local variables in an otherwise OO language is good.
I don't think we're talking about the same thing here. The distinction isn't between local and global variables at all. There are 2 types of local variables, and Python uses a third kind which doesn't really exist.
Static binding is meant to allow you to write (in Perl, say):
sub incr { my $n = shift; sub { $_[0]+$n } }
The python equivalent fails:
>>> def incr(n): ... return lambda x: x+n ... >>> f=incr(20) >>> f(10) Traceback (innermost last): File "", line 1, in ? File "", line 2, in NameError: n
Python's worst feature, in my opinion, is its half-baked approach to variable binding. Scheme and Lisp programmers have lexical binding (variables are looked up in the defining environment), and even Perl has this (`my' variables). Basic and Emacs Lisp programmers have dynamic binding (variables are looked up on the stack at runtime); Scheme, Lisp and Perl can use this (Perl via `local' variables).
Python has neither. Well, it has lexical binding, but it's "shallow", meaning the variables aren't looked up in surrounding environments. So you can have a closure, but you must explicitly import (via optional arguments, "n=n"!) variables from surrounding scopes. This is No Fun (tm).
Perl may have much syntactic ugliness, but at least it gets variable binding right.
> I block anything from Hotmail anyway, since only > spam ever comes from Hotmail, so who cares?
The last time I got spam from Hotmail, I sent an irrate letter to them. In reply, I got a very nice letter (sorry, don't have the person's name) explaining that all Hotmail mail gets an X-Originating-IP: header tacked on. So you can just filter on the existence of that line.
Anyone remember.Z files? gunzip can uncompress them, but gzip can't create them. That's because they use Unisys' state-of-the-art patented LZW algorithm. You need to use compress(1) to create them (people used to alias compress to a shorter name). Of course, the gzip algorithm gives (slightly) better results.
In fact, a recent/. poll showed that 0% of all users prefer.Z!
Burning nitrogen requires more energy than it gives (the triple bond is pretty strong). IIRC, the question was whether a nitrogen fusion reaction was sustainable in the atmosphere. It turned out not to be, so we had the nuclear arms race instead.
The problem with TLI is the content, not the presentation. While the universal grammar stuff seems to fit reasonably well to English, it starts falling apart (even to a layman such as myself) when considering Hebrew (my other language). This is particularly surprising in view of the fact that Chomsky knows Hebrew. I can only imagine what happens to the theory outside of its cosy English-Japanese-Hebrew playpen.
The popular anti-nativist work is _Educating Eve_, by Geoffrey Sampson. Unfortunately, both sides of this debate use very poor scientific arguments. The "argument from lack of data" (note the medievalist-sounding name) is a fine example: Nativists (Chomsky etc.) claim that there are insufficient data available to a child learning a language, particularly with regard to incorrectly-formed sentences (the child never hears someone being corrected for saying "the the the an orange chimpanzee", yet still knows it's wrong). Therefore, there's got to be some innate knowledge. The empiricists (anti-nativists, in this case) claim that grammar is the "simplest" explanation of the sentences presented to the child, so it gets selected.
But both sides present depressingly unscientific arguments! You'd think someone would go out and CHECK what children actually hear, but Sampson gives an example where Chomsky couldn't possibly have seen data for what he's claiming (creating the yes/no question for a sentence; you need to ask about the main verb, not about the first verb, and Chomsky claims that most children won't have heard a single example). Sampson gives examples of plausible sentences which would settle the matter. Of course, neither side has bothered to check it in the field.
But still, the empiricists seem to have it right:-)
Short names for "intrinsic" functions are good. It's not just the matter of preferring cos() to math_trig_cosine_angle(). Having all the string functions in the form strXXX() or strXXXX() lets you claim a portion of the namespace for the standard library (I think current C forbids application programs to use externally-scoped names of the form strXXX; perhaps you're allowed to have str_foo, but strfoo is definitely out).
Having compare_strings() would mean you're all over the namespace, and (as others have noted) string_compare() would fail due to the unique first 6 chars rule.
And, of course, common functions should have short names.
... but of course I'd guessed and was participating in the prank, which explains why I sent in a message DESIGNED to look as if I'd fallen for it, when IN FACT I'd realised that the older setup articles in Slashdot were just setting me up. I mean, you don't expect me to ADMIT to being a fool, do you?!
Now excuse me please while I go wipe the egg off my face.
Slashdot Mirror
In the spirit of automated classification and machine learning, I
tried searching for a set containing "boosting", "SVM" and "bagging"
(without "bagging", nothing new is found; another problem).
Results: "SVM", "Bagging", "Boosting", "stacking" and "Other methods".
Clicking on either of the 2 new links ("stacking" or "Other methods")
takes me to the normal Google search on the term. This is of course
not useful -- I need the terms in the machine learning context, which I
cannot get.
Especially in the case of "Other methods", it would be nice to be able
to get to the page Google had in mind...
That's implied by universality. Loosely, a CA is universal if it can compute (in some suitable encoding) any computable function (say, any function which can be computed by a Turing machine, or by a suitably unrestricted computer version of Fortran). The big brouhaha comes from the existence of CAs which are indeed universal (and Wolfram showed that many CAs are indeed universal).
Directly from the halting problem, e.g., you can get that it is uncomputable to decide, given a universal CA and an input configuration, whether the configuration "stabilizes" (remains unchanged after some iteration).
Rice's theorem says that it is uncomputable to decide, given a universal machine and any input (say, a universal CA and an initial configuration), anything interesting about the output.
Loosely speaking, "the only way to find out the final state of a cellular automaton is to run it".
I'm sure the FSF people have gone over this a million times (e.g. in gnu.misc.discuss), but here it is, again. The GPL doesn't force you to do anything. It only prohibits you from doing some things. Quoting from the article:
Just like any other software, you cannot redistribute GPL'd software except by certain license provisions. You cannot distribute derivative works of GPL'd software except by certain license provisions.
That is all. When you use Emacs, you aren't forced to do anything. When you sell (or otherwise give) Emacs to someone else, the GPL kicks in. If I download an e-book from somewhere (or buy a CD), I cannot distribute the data as I see fit -- I must follow the license under which I got it. Software is the same. GPL'd software is also no different.
As I always say, "if 2+2=5, then the Poincar Conjecture is true".
You're saying that there are observations in the real world that grossly disagree with relativity, but that you cannot replicate them experimentally.
Barring what goes on with the very small (quantum effects), what on earth are you talking about? I've never heard of any non-quantum contradictions to relativity. Of course quantum stuff works in the lab; could you please give references to such gross disagreements?
Note that the rapidly-spinning NASA antigravity is hardly claimed to be quantum, so those contradictions don't really help the NASA claimants, either.
That's not a problem, of course. Here's how you code in ARIELSOL: Unless you want to use the "UTM shortcut" (patent pending), just write out your program in ISO standard C. If you decide to take advantage of my NEW! UNIQUE! "UTM shortcut technology", your program should be the single bit `0' (note that this is not a legal C program).
The ARIELSOL interpreter (forthcoming) either compiles and runs your program as though it were a C program (if it's not the single bit 0), OR, if it is the single bit 0, runs itself on the input. Voila! Universality guaranteed, and in the same encoding.
I therefore claim that ARIELSOL encodes its universal machine in a single bit.
(I also claim that there is no Nobel prize in mathematics...)
I still don't get it. If I'm a disgruntled employee (say the company just collapsed and I've just been laid off and feel cheated), what's to stop me making a copy of any email to which I have access?
Saying "secure server" and "secure client" doesn't cut it. As long as I have reasonable access to my computer, I can make a copy. If the computer can display it for me to read, I can copy it.
Surely SeaLand protects against something else completely!
Tovarish, ITYM da.
(And don't believe the name, I can't speak Russian).
I know the "nabla" (aka "atled") symbol as "del", not the "partial" symbol.
..., depending how you use it
Idea for poll: correct name of upside-down triangle is:
a. upside-down triangle.
b. atled
c. nabla
d. del
e. grad / div /
That's one opinion. Another opinion is that Python supports only local variables, and that global variables are accepted as syntaxic sugar (useful for modules and functions defined out of classes). Some people think that allowing only local variables in an otherwise OO language is good.
I don't think we're talking about the same thing here. The distinction isn't between local and global variables at all. There are 2 types of local variables, and Python uses a third kind which doesn't really exist.
Static binding is meant to allow you to write (in Perl, say):
The python equivalent fails:
You need to pass n into the lambda explicitly.
Python's worst feature, in my opinion, is its half-baked approach to variable binding. Scheme and Lisp programmers have lexical binding (variables are looked up in the defining environment), and even Perl has this (`my' variables). Basic and Emacs Lisp programmers have dynamic binding (variables are looked up on the stack at runtime); Scheme, Lisp and Perl can use this (Perl via `local' variables).
Python has neither. Well, it has lexical binding, but it's "shallow", meaning the variables aren't looked up in surrounding environments. So you can have a closure, but you must explicitly import (via optional arguments, "n=n"!) variables from surrounding scopes. This is No Fun (tm).
Perl may have much syntactic ugliness, but at least it gets variable binding right.
> I block anything from Hotmail anyway, since only
> spam ever comes from Hotmail, so who cares?
The last time I got spam from Hotmail, I sent an irrate letter to them. In reply, I got a very nice letter (sorry, don't have the person's name) explaining that all Hotmail mail gets an X-Originating-IP: header tacked on. So you can just filter on the existence of that line.
Here's my procmail recipe which does just that:
:0 H:
* ^(From|X-From-Line|Return-Path):.*hotmail\.com
* !^X-Originating-IP:
junk
Anyone remember .Z files? gunzip can uncompress them, but gzip can't create them. That's because they use Unisys' state-of-the-art patented LZW algorithm. You need to use compress(1) to create them (people used to alias compress to a shorter name). Of course, the gzip algorithm gives (slightly) better results.
/. poll showed that 0% of all users prefer .Z!
In fact, a recent
OK, so one of the two links is a joke. Now all I have to do is figure out which one...
Burning nitrogen requires more energy than it gives (the triple bond is pretty strong). IIRC, the question was whether a nitrogen fusion reaction was sustainable in the atmosphere. It turned out not to be, so we had the nuclear arms race instead.
I really feel sorry for the poor little green men (tm) on Mars. Imagine being stuck with "embedded Windows NT" on every single 90GB hard disk!
Uhmm, isn't there a thing called "the slashdot effect" that happens when 1e11 people visiting slashdot click on the link?
The email message included 2 links. What would happen if *everyone* clicked on them?
We should make sure this doesn't happen.
The problem with TLI is the content, not the presentation. While the universal grammar stuff seems to fit reasonably well to English, it starts falling apart (even to a layman such as myself) when considering Hebrew (my other language). This is particularly surprising in view of the fact that Chomsky knows Hebrew. I can only imagine what happens to the theory outside of its cosy English-Japanese-Hebrew playpen.
:-)
The popular anti-nativist work is _Educating Eve_, by Geoffrey Sampson. Unfortunately, both sides of this debate use very poor scientific arguments. The "argument from lack of data" (note the medievalist-sounding name) is a fine example: Nativists (Chomsky etc.) claim that there are insufficient data available to a child learning a language, particularly with regard to incorrectly-formed sentences (the child never hears someone being corrected for saying "the the the an orange chimpanzee", yet still knows it's wrong). Therefore, there's got to be some innate knowledge. The empiricists (anti-nativists, in this case) claim that grammar is the "simplest" explanation of the sentences presented to the child, so it gets selected.
But both sides present depressingly unscientific arguments! You'd think someone would go out and CHECK what children actually hear, but Sampson gives an example where Chomsky couldn't possibly have seen data for what he's claiming (creating the yes/no question for a sentence; you need to ask about the main verb, not about the first verb, and Chomsky claims that most children won't have heard a single example). Sampson gives examples of plausible sentences which would settle the matter. Of course, neither side has bothered to check it in the field.
But still, the empiricists seem to have it right
Short names for "intrinsic" functions are good. It's not just the matter of preferring cos() to math_trig_cosine_angle(). Having all the string functions in the form strXXX() or strXXXX() lets you claim a portion of the namespace for the standard library (I think current C forbids application programs to use externally-scoped names of the form strXXX; perhaps you're allowed to have str_foo, but strfoo is definitely out).
Having compare_strings() would mean you're all over the namespace, and (as others have noted) string_compare() would fail due to the unique first 6 chars rule.
And, of course, common functions should have short names.
... but of course I'd guessed and was participating in the prank, which explains why I sent in a message DESIGNED to look as if I'd fallen for it, when IN FACT I'd realised that the older setup articles in Slashdot were just setting me up. I mean, you don't expect me to ADMIT to being a fool, do you?!
Now excuse me please while I go wipe the egg off my face.