My point was not that floating point are better or worse. My point is that 3 digit accuracy on the result IS NOT 3 digit accuracy in the computation.
This effect is known and dealt with in processor by having FPU registers bigger than CPU registers so that internal computation got more accuracy
Moreover, when you do some image filtering, you do not know in advance whether you are going to completely get rid of a color component or not. It is thus difficult to estimate how much precision you need. This is the point of choosing float or doubles. You suppose that a xx-bit floating point number would be enough for your computations.
Lowering the accuracy of floating point computation to get more power is not new. The cell processor used to do that...
So this technology could be useful if you could tune your precision requirement. But I do not believe that you could really tune it efficiently from the hardware.
There's a whole lot we are calculating now without the need for more than 3 significant digits - and a whole bunch where we intentionally use random numbers, sometimes even with strong hardware entropy gathering.
mmm, yes you generally don't need more than 3 significant digit on the final result. But all the intermediate computation need to be more accurate to reach a 3 digit accuracy at the end. Think about a computation such as X - Y. If X and Y have the 3 first same digit, then you will have 0 significant bit for the result.
if you do anything more complicated than adding and multiplying, then you need accurate computation.
Honestly, I think I haven't yet seen someone receive comments back where they couldn't take a good guess at who the originator was...
Most of the time, I do not know who read my articles and I do not believe the review I wrote tells who I am. There is a bunch of "top-level" researcher in my field (parallel computing) and probably 10 times more "classical researcher" and PhD student. Perhaps it is not true in all field.
Even if your identity can be guessed, it is never entirely sure. And being anonymous prevent pressure such as "you rejected my last article, I will reject your". You won't risk it since you are not sure it is the same guy. And the editor knows you, you will lose credibility to him if you are proven wrong. And being proven wrong can be easy when the authors can answer to reviews.
In short, anonymity is a major property in the scientific peer reviewing process.
In Computer Sciences, there are several journal and conferences where the reviewing process is double blind.
Researchers that got more than one paper per week generally push the review to PhD student. Moreover, most of the time reviews are weighted with a confidence index so that reviewers can tell the editor 'it sounds correct but I did not checked in details'.
well. linux is not that bad. Probably they will not only tweak linus but also develop what they are interested in. What is the point of developing something else. Could they really do in a reasonable amount of time something better than linux ? I mean in term of driver support, compatibility etc. I believe you can not easily do better. So they team up with other people which seems to be the best idea.
( Oh, they could have used hurd )
No they are not.
So it is possible that their papers have been cited after being nobel prize.
From what i understood of the paper, this index is basically weighted citation index which considers how many references do an article cite and recursively (with exponentially decreasing weight).
A little awk goes a long way. Not the big-bad-I-am-a-programming-language-awk, but the smaller-friendlier extract one or two columns of text from something awk. ex. awk '{print $2}' prints the thing in the second column. Add -F the field separator tool and it gets really useful.
Better example. Here is a postfix log line.
I do this kind of thing with "cut" which is very useful when used with "paste" and "join"
...is to tell people which paper they should cite.
People will cite something when referencing your software. Because it is required and because there is no reason NOT to cite your paper.
However, sometimes, you don't know which paper to cite. To avoid this, you should tell people which of your papers describes best your software. Moreover, it will concentrate citation on a single paper which is usually better.
I have only a french edition of the K&R second edition. Thus the following is a translation from me and could be inconsistent with the english version.
In section 2.12 "priority and evolution order" is written:
The language does no specify the order of evaluation of the arguments of a function. Thus
"printf ("%d %d\n", ++n, pow(2,n));" can lead to different results depending on the compiler.
Later in the same section is written:
The instruction a[i] = i++; can be dangerous. The problem is to know whether the index uses the old or the new value of i. Compilers can interpret this in different manners, leading to various results depending on their understanding of the expressions
I heard a rumor that Google will implement their own version of ad block, which won't fully stop the ads from loading, but rather will hide them in the rendering process. Advertisers still see an ad generate, and users don't see the ad.
I heard that General Motors will make better motors and will push towards bikes.
The web is about datas and I am a bit upset that my webbrowser has been taken over by web designer.
Why can not I choose the way datas are displayed ?
Web designer does not know how I would like datas to be displayed on my screen. They do not know the resolution of my window, how many color my screen can display or its refresh rate. I could browse using e-ink technology and thus animations would be wrong.
They do not even know if i got a screen or a braille reader or an audio webbrowser.
I dream every day of an Internet with no web. I mean, why all those data in slashdot are displayed like that ? Why can't I write my own software to display it ? I am fine with the web master of/. to provide me ONE way of displaying it. But why can't i chose everything by myself.
The web is about datas. Not about how they are to be displayed.
godrik@mandan:tmp$ cat test.c
int fnct (int i)
{
return ++i + ++i;
}
godrik@mandan:tmp$ splint test.c
Splint 3.1.1 --- 03 Nov 2006
test.c: (in function fnct)
test.c:3:16: Expression has undefined behavior (left operand uses i, modified
by right operand): ++i + ++i
Code has unspecified behavior. Order of evaluation of function parameters or
subexpressions is not defined, so if a value is used and modified in
different places not separated by a sequence point constraining evaluation
order, then the result of the expression is unspecified. (Use -evalorder to
inhibit warning)
test.c:3:16: Expression has undefined behavior (left operand modifies i, used
by right operand): ++i + ++i
Finished checking --- 2 code warnings
yes, CPU clock speeds are going up, and memory prices are going down, but a web browser should still be a relatively lightweight application by itself.
Why? I spend more time in the web browser--by far--than any other application. Email? 10 years ago I used a standalone email app, now I mostly use webmail. 5 years ago I used AIM. Now I use web chat. Picasa? Google documents? Between javascript advances, DOM, rich media, plugins, TABS, etc etc etc, today's browser does things not even imaged in 10 years ago.
Chrome's very purpose is to make the browser a more generalized application development platform. Heck, WEBKIT is used in the same way (and XUL, etc for Firefox). The web browser ain't just for HTML circa '97 anymore. The web browser is probably the single most important application for most users.
People using web application != people should use web application.
Web application are pain in the ass.
Everthing, i mean EVERYTING people do with there 2.7Ghz dual core 2 GB memory computer today, i did it in 1997. I checked my mail using outlook. Browse the web for informations on IE. Streaming musics. Streaming videos (not that much considering the bandwidth). Reading to forums and newsgroups.
Well, basically, people do the same thing today. They uses at least ten time more ressource but do the same thing ? Why should we say. Because everything is done one the web, because "It's so cool". Don't tell me it is because web applications are easier to use. They are not. Look at yahoo mail, then look at outlook express. It is the same application.
Why use flashplayers to stream some videos ? mplayer can do that.
There is no interest in this so called web 2.0. No interest at all unless you want to use more and more ressources.
Slashdot Mirror
My point was not that floating point are better or worse. My point is that 3 digit accuracy on the result IS NOT 3 digit accuracy in the computation.
This effect is known and dealt with in processor by having FPU registers bigger than CPU registers so that internal computation got more accuracy
Moreover, when you do some image filtering, you do not know in advance whether you are going to completely get rid of a color component or not. It is thus difficult to estimate how much precision you need. This is the point of choosing float or doubles. You suppose that a xx-bit floating point number would be enough for your computations.
Lowering the accuracy of floating point computation to get more power is not new. The cell processor used to do that...
So this technology could be useful if you could tune your precision requirement. But I do not believe that you could really tune it efficiently from the hardware.
There's a whole lot we are calculating now without the need for more than 3 significant digits - and a whole bunch where we intentionally use random numbers, sometimes even with strong hardware entropy gathering.
mmm, yes you generally don't need more than 3 significant digit on the final result. But all the intermediate computation need to be more accurate to reach a 3 digit accuracy at the end. Think about a computation such as X - Y. If X and Y have the 3 first same digit, then you will have 0 significant bit for the result.
if you do anything more complicated than adding and multiplying, then you need accurate computation.
Honestly, I think I haven't yet seen someone receive comments back where they couldn't take a good guess at who the originator was...
Most of the time, I do not know who read my articles and I do not believe the review I wrote tells who I am. There is a bunch of "top-level" researcher in my field (parallel computing) and probably 10 times more "classical researcher" and PhD student. Perhaps it is not true in all field.
Even if your identity can be guessed, it is never entirely sure. And being anonymous prevent pressure such as "you rejected my last article, I will reject your". You won't risk it since you are not sure it is the same guy. And the editor knows you, you will lose credibility to him if you are proven wrong. And being proven wrong can be easy when the authors can answer to reviews.
In short, anonymity is a major property in the scientific peer reviewing process.
In Computer Sciences, there are several journal and conferences where the reviewing process is double blind.
Researchers that got more than one paper per week generally push the review to PhD student. Moreover, most of the time reviews are weighted with a confidence index so that reviewers can tell the editor 'it sounds correct but I did not checked in details'.
drawing smiley on google maps with my friends and our GPS
well. linux is not that bad. Probably they will not only tweak linus but also develop what they are interested in. What is the point of developing something else. Could they really do in a reasonable amount of time something better than linux ? I mean in term of driver support, compatibility etc. I believe you can not easily do better. So they team up with other people which seems to be the best idea. ( Oh, they could have used hurd )
The question should be. What can the GPL do against a state. What would happen if they decide to develop and distribute a closed source version ?
No they are not. So it is possible that their papers have been cited after being nobel prize. From what i understood of the paper, this index is basically weighted citation index which considers how many references do an article cite and recursively (with exponentially decreasing weight).
A little awk goes a long way. Not the big-bad-I-am-a-programming-language-awk, but the smaller-friendlier extract one or two columns of text from something awk. ex. awk '{print $2}' prints the thing in the second column. Add -F the field separator tool and it gets really useful. Better example. Here is a postfix log line.
I do this kind of thing with "cut" which is very useful when used with "paste" and "join"
...is to tell people which paper they should cite. People will cite something when referencing your software. Because it is required and because there is no reason NOT to cite your paper.
However, sometimes, you don't know which paper to cite. To avoid this, you should tell people which of your papers describes best your software. Moreover, it will concentrate citation on a single paper which is usually better.
You mean someone who understood that the only way to make windows work flawlessly is to draw pentagrams and to sacrifice chickens ?
Well, I can do that
well, it will probably be written in latex. So it will be compiled to pdf.
Moreover, open-source is sort of a buzz word
I have only a french edition of the K&R second edition. Thus the following is a translation from me and could be inconsistent with the english version.
In section 2.12 "priority and evolution order" is written:
The language does no specify the order of evaluation of the arguments of a function. Thus "printf ("%d %d\n", ++n, pow(2,n));" can lead to different results depending on the compiler.
Later in the same section is written:
The instruction a[i] = i++; can be dangerous. The problem is to know whether the index uses the old or the new value of i. Compilers can interpret this in different manners, leading to various results depending on their understanding of the expressions
I heard a rumor that Google will implement their own version of ad block, which won't fully stop the ads from loading, but rather will hide them in the rendering process. Advertisers still see an ad generate, and users don't see the ad.
I heard that General Motors will make better motors and will push towards bikes.
The web is about datas and I am a bit upset that my webbrowser has been taken over by web designer.
Why can not I choose the way datas are displayed ?
Web designer does not know how I would like datas to be displayed on my screen. They do not know the resolution of my window, how many color my screen can display or its refresh rate. I could browse using e-ink technology and thus animations would be wrong.
They do not even know if i got a screen or a braille reader or an audio webbrowser.
I dream every day of an Internet with no web. I mean, why all those data in slashdot are displayed like that ? Why can't I write my own software to display it ? I am fine with the web master of /. to provide me ONE way of displaying it. But why can't i chose everything by myself.
The web is about datas. Not about how they are to be displayed.
godrik@mandan:tmp$ cat test.c int fnct (int i) { return ++i + ++i; } godrik@mandan:tmp$ splint test.c Splint 3.1.1 --- 03 Nov 2006 test.c: (in function fnct) test.c:3:16: Expression has undefined behavior (left operand uses i, modified by right operand): ++i + ++i Code has unspecified behavior. Order of evaluation of function parameters or subexpressions is not defined, so if a value is used and modified in different places not separated by a sequence point constraining evaluation order, then the result of the expression is unspecified. (Use -evalorder to inhibit warning) test.c:3:16: Expression has undefined behavior (left operand modifies i, used by right operand): ++i + ++i Finished checking --- 2 code warnings
yes, CPU clock speeds are going up, and memory prices are going down, but a web browser should still be a relatively lightweight application by itself.
Why? I spend more time in the web browser--by far--than any other application. Email? 10 years ago I used a standalone email app, now I mostly use webmail. 5 years ago I used AIM. Now I use web chat. Picasa? Google documents? Between javascript advances, DOM, rich media, plugins, TABS, etc etc etc, today's browser does things not even imaged in 10 years ago.
Chrome's very purpose is to make the browser a more generalized application development platform. Heck, WEBKIT is used in the same way (and XUL, etc for Firefox). The web browser ain't just for HTML circa '97 anymore. The web browser is probably the single most important application for most users.
People using web application != people should use web application. Web application are pain in the ass. Everthing, i mean EVERYTING people do with there 2.7Ghz dual core 2 GB memory computer today, i did it in 1997. I checked my mail using outlook. Browse the web for informations on IE. Streaming musics. Streaming videos (not that much considering the bandwidth). Reading to forums and newsgroups. Well, basically, people do the same thing today. They uses at least ten time more ressource but do the same thing ? Why should we say. Because everything is done one the web, because "It's so cool". Don't tell me it is because web applications are easier to use. They are not. Look at yahoo mail, then look at outlook express. It is the same application. Why use flashplayers to stream some videos ? mplayer can do that. There is no interest in this so called web 2.0. No interest at all unless you want to use more and more ressources.
Oh, this will surely help color-blind peoples!