sgolayfilt no longer support SINGLE
in [Matlab]
|
From: Jan Simon on 11 Jun 2010 19:26 Dear Sscnekro, > > Sorry, I missed something: Why do you try to contact Jack Little? > Why not, on earth? Anybody can do. In this event, it's a trial *for fun* to see what it gives. His reaction or lack of reaction will say again something on TMW. > You may take it also that way: Suppose you 'd be the founder, with the vison-mission-ambition we already know. Would there be a possibility to contact you? 2nd guideline on the top of this form for posting: "Specific questions elicit the best response from this unmoderated newsgroup." Do you have a specific question to Jack? Do I understand correctly, that you want to ask if TMW uses standardized feedback mechanisms? If so, it might be possible that Jack does not answer, because he has another job in this big company. BTW.: @Steve: thanks for joining this thread. @Bruno: Sorry that SGOLAYFILT went out of view. Actually, I just wanted to mention that I've added support for SINGLE data in fSGolayFilt. Casting input and output of SGOLAYFILT is not efficient. fSGolayFilt accumulates the dot products in a DOUBLE to encrease accuracy. Perhaps this is of any need for you. Jan
From: dpb on 11 Jun 2010 19:32 Steve Eddins wrote: > On 6/11/2010 9:35 AM, dpb wrote: .... >> A recasting of the ideas in somewhat more modest terms I could >> stomach...that wording seems "over the top". .... > I understand your reaction. I've also been skeptical of some corporate > mission statements. But I can connect every phrase of the technology, > business, human, and social mission statements to concrete and > long-lasting attitudes, activities, and practices in place at MathWorks. > The joy and reward of striving to achieve these challenging ideals is > what keeps me coming to work. .... On reflection I have to admit much of my antipathy stems from the point some years ago when I last tried to upgrade on the eve of seeing retirement looming but thinking it would be good to have a more current version of ML if wanted to continue to participate at cssm even though could foresee no financial benefits accruing therefrom. To cut to the chase, I found little to reflect the "human or social mission" in the response but a lot of business. Now, I fully understand it takes money to run a company and I'm not begrudging TMW its success but they didn't endear themselves to me or strike me as being interested in being even the least bit altruistic. --
From: Joaquim Luis on 11 Jun 2010 20:39 "Steven Lord" <slord(a)mathworks.com> wrote in message <hutghu$6rt$1(a)fred.mathworks.com>... > > "Joaquim Luis" <jluis@--ualg--.pt> wrote in message > news:hurt6g$eu3$1(a)fred.mathworks.com... > > Honglei Chen <Honglei.Chen(a)mathworks.com> wrote in message > > <hq2pim$599$1(a)fred.mathworks.com>... > >> Hi Bruno, > >> > >> We take backwards compatibility very seriously. > > > > You do??? > > So may I ask, where is the Maltab compiler right now? > > Right here. > > http://www.mathworks.com/products/compiler/ > > If you intended to ask "Why doesn't MATLAB Compiler generate C or C++ code > that we can modify like previous versions did?" there are a number of > reasons that I don't want to get into (and I'm not sure I am allowed to get > into.) > > If you want to get technical, EACH AND EVERY single enhancement we make to > our products and ALMOST EVERY single bug fix we make are incompatibilities. > If we never wanted to introduce any incompatibilties, we could say "Okay, > our products are done", much like Donald Knuth has requested that TeX and > METAFONT remain unchanged after his (hopefully far in the future) death: > > http://en.wikipedia.org/wiki/TeX > > But I'd like to think that our users would prefer that we try to make our > products better and fix bugs and introduce new features they request. > [Since I _know_ you're going to ask how the newer versions of MATLAB > Compiler are better than the older versions: the newer versions can, among > other things, compile MATLAB classes -- the older versions could not do so. > I'm almost certain that was THE #1 requested feature for the Compiler (by a > LARGE margin) before we introduced that capability in MATLAB Compiler 4.0.] No, I did not intend to ask that because I know those arguments for long time and, in my view, I don't accept them as valid. Why so? Because they are not answers to the question. The replacement product that abusively took the the name of "compiler" certainly has its own merits. No dispute on that. But it is NOT A COMPILER ans it's not a replacement of the compiler. With the compiler, and all its limitations that could have been reduced with continuous support, one could build reasonable sized executables that loaded in 1 second. With the encrypter, well be very patience with those couple hundred Mb one need to ship on even for a very small compiled code, which is not compatible between releases, etc ...
From: Bruno Luong on 12 Jun 2010 05:13 "Jan Simon" <matlab.THIS_YEAR(a)nMINUSsimon.de> wrote in message <huugmd$acn$1(a)fred.mathworks.com>... > > @Bruno: Sorry that SGOLAYFILT went out of view. Actually, I just wanted to mention that I've added support for SINGLE data in fSGolayFilt. Casting input and output of SGOLAYFILT is not efficient. fSGolayFilt accumulates the dot products in a DOUBLE to encrease accuracy. Perhaps this is of any need for you. I have yet to test your fSGolayFilt Jan. I must first to make sure the stability before I can decide to replace Matlab SGOLAYFILT. It is very easy to make a work around for Matlab SGOLAYFILT, edit the mfile 1. Comment out line 51-55 % try % chkinputdatatype(x,k,F,W,DIM); % catch ME % throwAsCaller(ME); % end 2. Before Line 84 add the cast of the Filter definition: B = feval(class(x),B); ytemp = filter(B((F-1)./2+1,:),1,x); Now the function can work on DOUBLE and SINGLE without casting the big array. That's a much better workaround than casting the large data array "x" single to 'double" as Honglei Chen has suggested. It is quite understandable that Mathworks would prefer to give a workaround on the user side rather than modifying what's considered as "completely finished and released" - for the sake of quoting Steve Eddins words. What frustrate me is Mathwork in one hand they advertises the extension of the support of SINGLE in signal processing toolbox, and on the other hand restrict the flexibility of some code by brute type-checking. This seems to me represents a not well coordinated development plan. The issue here seems to be an isolate case, and I hope it is. Bruno
From: sscnekro on 12 Jun 2010 07:59
> Do you have a specific question to Jack? Do I understand correctly, that you want to ask if TMW uses standardized feedback mechanisms? If so, it might be possible that Jack does not answer, because he has another job in this big company. Never mind, Jan, you will probably never understand this sort of different mentality than is yours. In your micro-view, things have got to be as they are, bcs somebody else said it has got to be like that. You might just learn that there are different micro-views. You may take this as testing, for fun, how much is enough to be interesting. $*10^6? $*10^10? More? You may test the hypothesis that posters to raise technical issues under this thread are crumbs that is not interesting to collect. Maybe the case of a social networking site is familiar to you, telling people that things have got to be as they are. At a point >3*10^4 people left the network they admitted that things yet might have got to be a bit different. This could hardly happen here as you seem to experience self-reproaches and uselesness from any bit of questions that enter your thoughts. Now, pssssst, let us not disturb this hard work for this big company. (Don't understand me wrongly, I am not disparaging the achievments and success of TMW.) |