MATLAB Central Newsreader Recent Posts

Re: optimization problem: get a feasible point for running an optimization

Torsten — Mon, 04 Jun 2012 09:40:01 -0400

>
> I used many times fmincon with a not feasible starting point and I got an
> unfeasible solution. I read on a matlab documentation, fmincon needs a feasible starting point, therefore I was wondering it was possible to find one througn the optimization toolbox.

Depends on the kind of constraints that you have.
E.g. if they are all linear, you could use LINPROG to find a feasible
starting point.

Best wishes
Torsten.

How to retain the values in "inputdlg"?

Prashant Sandhi — Mon, 04 Jun 2012 09:36:07 -0400

Dear All,

I have a code where I can enter the values and after pressing 'OK' then all the values are loaded into 'workspace'. Imagine if I want to change the values of 'N' & 'CWL' later, then in my below code I have to enter all the values again once I press 'OK'.

*******************************************************************
clear all;clc;
Data=inputdlg( {'Enter Sampling Frequency in Hz:','Enter Cutoff Frequency in Hz:',...
'Enter Order of the Filter:','Enter Input Word Length 8bit, 12bit,..32bit..:',...
'Enter Coefficient Word Length:'});

% All frequency values are in Hz.
Fs = str2num(Data{1}); % Sampling Frequency
Fc = str2num(Data{2}); % Cutoff Frequency
N = str2num(Data{3}); % Order of the Filter
n = str2num(Data{4}); % Enter Input Word Length
CWL = str2num(Data{5}); % Enter Coefficient Word Length

********************************************************************
To avoid entering all the values each time in the input-dialogue-box, I would like to perform two things:
1. Let the 'inputdlg' box be open even after pressing 'OK'
2. Retain all the previous values I have entered in 'inputdlf' box.

This will avoid entering the values for all Fs, Fc, N, n & CWL.

Any suggestions will be appreciated.

Thanking you in advance!

Re: optimization problem: get a feasible point for running an optimization

pietro — Mon, 04 Jun 2012 09:28:09 -0400

"Seth Deland" <sdeland@mathworks.com> wrote in message <jqaa7l$m85$1@newscl01ah.mathworks.com>...
> To add to what Sargondjani said, if the solver is having trouble finding any
> feasible point, there is this section of the documentation which provides
> some guidance:
> http://www.mathworks.com/help/toolbox/optim/ug/br44iv5-1.html#br44i73
>
>
> "Sargondjani " <sponsstimo@hotmail.com> wrote in message
> news:jqa682$76k$1@newscl01ah.mathworks.com...
> > i dont think it needs a feasible point to start... i mean, during
> > iterations it can also violate constraints, so i dont see why your initial
> > guess should be feasible...

Thanks for yout reply,

I used many times fmincon with a not feasible starting point and I got an
unfeasible solution. I read on a matlab documentation, fmincon needs a feasible starting point, therefore I was wondering it was possible to find one througn the optimization toolbox.

Re: Capturing Image from built in webcam

Aqeel Anwar — Mon, 04 Jun 2012 09:19:09 -0400

"Image Analyst" wrote in message <jqgpoq$85t$1@newscl01ah.mathworks.com>...
> Aqeel Anwar:
> You'll probably have to find some API or SDK code for your web cam. Then use loadlibrary() and make calls directly to that.

Actually I am a beginner to MATLAB, so can you please help me finding one?

Re: 4D matrix manipulation

Matt J — Mon, 04 Jun 2012 09:12:06 -0400

"Matt J" wrote in message <jqht3f$n50$1@newscl01ah.mathworks.com>...
> "Niels " <nrvdwerf@gmail.com> wrote in message <jqhsq3$m3c$1@newscl01ah.mathworks.com>...
> >
> >
> > Can you please help me find a short algorithm to fill the TAC matrix?
> ==========
>
> TAC=reshape(X,[],63).';

Make that

TAC=reshape(X,[],19).';

Re: 4D matrix manipulation

Matt J — Mon, 04 Jun 2012 08:54:08 -0400

"Niels " <nrvdwerf@gmail.com> wrote in message <jqhsq3$m3c$1@newscl01ah.mathworks.com>...
>
>
> Can you please help me find a short algorithm to fill the TAC matrix?
==========

TAC=reshape(X,[],63).';

Re: fmincon: convergence and choice of algorithm

Matt J — Mon, 04 Jun 2012 08:50:07 -0400

"Olivier" wrote in message <jqhpqe$auq$1@newscl01ah.mathworks.com>...
>
> tempCov is simply a (annualized) covariance Matrix (70x70). I can post this covariance matrix if this is helpful.
>
> My observation is the following:
> 1) the convergence is very slow. I need to increase the exit options to optimset(... 'TolX',1e-8,'TolFun',1e-8,'MaxIter',100000, 'MaxFunEvals', 100000) to get a result which is "close enough" to the close formula.
>
> I think this is a "relevant" observation. Many papers in finance solve this kind of problems (convex optimization problem in 70 dimensions). They generally solve the problem with sqp but there is no indication about the "precision".
>
> I have to correct my other observation:
> 1) the sqp and interior-point algorithms have very similar solutions but not exactly the same solution
=====================

But how fast is QUADPROG? I haven't read papers in the field of finance, but I assume that this is what they would normally be using, since it's the best suited for the quadratic structure of your problem.

And how well conditioned is your covariance matrix, i.e. what is cond(tempCov)?
A poorly conditioned covariance matrix could certainly influence speed. Although fmincon tries to estimate the covariance and thereby compensate for ill-conditioning, it would be harder for it to do so when you haven't provided it the gradient and Hessian explicitly, using the GradObj and Hessian options.

Finally, certain fmincon algorithms approximate the Hessian (the covariance in your case) with different levels of precision. The sqp algorithm, which you are using for example, doesn't give you an option to supply the Hessian explicitly, but rather estimates it on the fly.

4D matrix manipulation

Niels — Mon, 04 Jun 2012 08:49:07 -0400

Hello!

For my bachelor thesis I have to manipulate a 4D matrix (X). The first three dimensions are the x,y,z directions and the fourth dimension is time.

What I need from this data is the time dependency at every point, which I like to store in a 2D matrix called TAC (time activity).

I already now what to do for the first coordinate (1,1,1,), namely:

TAC(:,1) = X(1,1,1,:)

The lengths of the different dimensions are:
length_x = 128
length_y = 128
length_z = 63
length_x = 19

So, the TAC matrix will be a 19 x (128*128*63) = 19 x 1032192 matrix
And it's first row will be the time activity of coordinate (1,1,1). The second row should contain the time activity of coordinate (2,1,1) etc.

Can you please help me find a short algorithm to fill the TAC matrix?

Re: Accessing Non-Excel Libre Spreadsheet from MATLAB

mklcst mklcst — Mon, 04 Jun 2012 08:46:06 -0400

TideMan <mulgor@gmail.com> wrote in message <14349772.98.1332892357752.JavaMail.geo-discussion-forums@ynaz5>...
> On Wednesday, March 28, 2012 11:53:14 AM UTC+13, Eric Diaz wrote:
> > > When running XLSREAD on a machine that is not a Windows machine with
> > > Microsoft Excel installed, XLSREAD will operate in "basic mode":
> > >
> > > http://www.mathworks.com/help/techdoc/ref/xlsread.html
> > >
> > > "[num,txt,raw] = xlsread(filename,sheet,range,'basic') reads data from the
> > > spreadsheet in basic mode, the default on systems without Excel for
> > > Windows."
> > >
> > > Basic mode has some limitations, as described in the Input Arguments section
> > > of that documentation, and it appears you were trying to do something that
> > > exceeded the capabilities of basic mode.
> > >
> >
> > Right...I remember it saying something about "basic mode" too, but it still gave the errors.
> >
> > > > I don't really understand this COM server stuff at all? Is there a COM
> > > > server method for accessing open-source office spreadsheets?
> > >
> > > If the program offers a COM interface, yes, but a quick glance at the
> > > LibreOffice website didn't indicate that it had such an interface.
> >
> > OK, thanks. I guess that is about as close to an answer as I can get, since you work at TMW. Do you know whether there is any way to resolve this problem other than bowing down to the almighty Microsoft Office? I would definitely like to see support for an open office suite, like Libre.
>
> A work around would be to save the spreadsheet as a .csv file, then getting the data into Matlab is a breeze (and faster than xlsread as well).

I think it's time for Matlab to fully support libreoffice (or OpenOffice or whatelse).
A part for the office license, it would amazing also for the unix platform linux and Mac.

Re: fmincon: convergence and choice of algorithm

Torsten — Mon, 04 Jun 2012 08:20:27 -0400

Forget my reply - it's nonsense.

Best wishes
Torsten.

Re: fmincon: convergence and choice of algorithm

Torsten — Mon, 04 Jun 2012 08:18:34 -0400

Does fmincon accept your setting although Aeq is a vector instead of a
matrix and beq is a scalar instead of a vector ?

Best wishes
Torsten.

fmincon: convergence and choice of algorithm

Olivier — Mon, 04 Jun 2012 07:58:06 -0400

Hi,

A couple of days ago, I posted already a question regarding the convergence of fmincon. Let me make it a bit more concrete.

I solve the following problem with fmincon:

x0=ones(N,1) ./(N);
    f=@(x)funMV(x,tempCov);
    A = [];
    b = [];
    Aeq = ones(1,N);
    beq = 1;
    lb =-inf * ones(N,1);
    %lb = zeros(N,1);
    ub = ones(N,1).*inf;
    options = optimset('Algorithm', 'sqp', 'Display', 'off', 'TolX',1e-6,'TolFun',1e-6,'MaxIter',10000, 'MaxFunEvals', 10000);
    MV = fmincon(f,x0,A,b,Aeq,beq,lb,ub,[],options);

with:
       function y = funMV(x,tempCov)
            y = x' * tempCov * x;
        end

This problem is finding the minimum variance portfolio and is well described in the literatur. I know that we can solve it with a close formula (or with quadprog). I do this exercice to see how "good" different solvers are.

tempCov is simply a (annualized) covariance Matrix (70x70). I can post this covariance matrix if this is helpful.

My observation is the following:
1) the convergence is very slow. I need to increase the exit options to optimset(... 'TolX',1e-8,'TolFun',1e-8,'MaxIter',100000, 'MaxFunEvals', 100000) to get a result which is "close enough" to the close formula.

I think this is a "relevant" observation. Many papers in finance solve this kind of problems (convex optimization problem in 70 dimensions). They generally solve the problem with sqp but there is no indication about the "precision".

I have to correct my other observation:
1) the sqp and interior-point algorithms have very similar solutions but not exactly the same solution

Re: Trying to use extended ASCII code in Matlab

Christoph — Mon, 04 Jun 2012 07:37:09 -0400

"Star Strider" wrote in message <jqb2p7$ga6$1@newscl01ah.mathworks.com>...
>
> It seems this is not a problem with MATLAB. The ‘Help’ for ‘char’ says:
>
> ‘S = char(X) converts array X of nonnegative integer codes into a character array. Valid codes range from 0 to 65535, where codes 0 through 127 correspond to 7-bit ASCII characters. The characters that MATLAB can process (other than 7-bit ASCII characters) depend upon your current locale setting. To convert characters into a numeric array, use the double function.’
>
> If you have a Windows system, open ‘Character Map’ and see what it displays. On mine (Windows 7, U.S.), it apparently defaults to ‘Unicode’ without giving any further information, so char(252) yields ‘u’ with an umlaut, and double('ü') yields 252.0000. ASCII beyond 127 doesn’t seem to be an option because Windows doesn’t support it.

Hi Star,

thanks for your reply. I read the help text, too. As far as I know the encoding of the character depends not on the local system, but on the given font. Furthermore ACSII beyond 127 is not defiend because ASCII is, as you said before, just a 7bit code. The ASCII character 128 up to 255 are not defined distinctly.

Nevermind, I tried to get character 2560 in arial (looked-up in charactermap on windows), but with char(2560) I just get a not displayed character, which implies that Matlab uses a different encoder or does not support this character. I guess the second might be realy poor, because this kind of characters allows you to draw simple paths without using any grafical interface but strings.

If anybody has an idea how to get the "Rahmenelement" which might be "Frame-Element" in English feel free to give me a note.

so long,
CN

So long,
CN

Questions > Warning: Executing startup failed in matlabrc.

Kelvin — Mon, 04 Jun 2012 07:34:09 -0400

Hi all pros,

I'm trying to use timer and startat to schedule some daily and monthly process to automatically generate reports and stuff, so I created a startup file with the following code and put it under D:\MATLAB TESTSITE

However, whenever Matlab opens, the following error message is shown, so I was wondering if anyone knows what went wrong and how can I fix it?

Thank you in advance.

P.S.
I'm trying to generate 2 reports on daily basis at 9:00 and 21:00 and update database on monthly basis on 10th, 20th and 30th (on 28th if it is February)

%%%%%%%%%%%%%%%%%%%%%%%%%%%
Warning: Executing startup failed in matlabrc.
This indicates a potentially serious problem in your MATLAB setup,
which should be resolved as soon as possible. Error detected was:
MATLAB:timer:startdelaynegative
The specified start time has already passed.
%%%%%%%%%%%%%%%%%%%%%%%%%%%

cd 'D:\MATLAB TESTSITE'
t = timer('TimerFcn', 'MonthlyEarlyRun');
startat(t, '21:00:00');

cd 'D:\MATLAB TESTSITE'
t1 = timer('TimerFcn', 'DayRun');
startat(t1,'09:00:00');

[y, m, d, h, mn, s] = datevec(today);

if d == 30
    cd 'D:\MATLAB TESTSITE'
    t2 = timer('TimerFcn', 'MonthlyEarlyRun');
    startat(t2,'12:00:00');
end

if d == 20
    cd 'D:\MATLAB TESTSITE'
    t3 = timer('TimerFcn', 'MonthlyMiddleRun');
    startat(t3,'12:00:00');
end

if d == 30
    cd 'D:\MATLAB TESTSITE'
    t4 = timer('TimerFcn', 'MonthlyLateRun');
    startat(t4,'12:00:00');
end

if m == 2 && d == 28
    cd 'D:\MATLAB TESTSITE'
    t5 = timer('TimerFcn', 'MonthlyLateRun');
    startat(t5,'12:00:00');
end

Re: Correct use of fftshift and ifftshift at inpu tto fft and ifft

Du Huynh — Mon, 04 Jun 2012 07:09:06 -0400

I always avoid to use fftshift and ifftshift as much as I can. If you pass a 2D image to the fft2 function, then the output 2D array always has the DC term at the first element position (i.e., the top-left corner of the 2D array for the row-column coordinate system). If all the frequency-domain filters that you design have the DC terms at the same position, then you don't need to use fftshift or ifftshift. I would use fftshift only if I want to visualize the Fourier spectrum of my image.

I have verified the following from my frequent uses of these functions:
* fftshift = ifftshift if your image has even numbers of rows and columns.
* always use fftshift if you want to move the DC term from the top-left corner to the centre of the 2D array, regardless of the size of your input image (odd or even number of rows and colums).
* always use ifftshift if you want to move the DC term from the centre of the 2D array to the top-left corner, regardless of the image size.
* label your frequency axes correctly. If the number of elements is odd (i.e. N is odd below), then use (assuming that your sampling rate is 1)
w = linspace(-0.5, 0.5, N)
as your frequency array.
If the number of elements is even (i.e. N is even below), then use
w = linspace(-0.5, 0.5, N); w(end) = [ ]
as your frequency array.

Very often we found that it is easier to directly design a filter in the frequency domain with the DC term in the centre of the array. If this is the case, then I would suggest that, after you have got your filter, you call ifftshift to swap the quadrants so that the DC term is at the top-left corner of the image. From then on, everything you operate on has the DC terms at the top-left corner. To move back to the spatial domain, you don't need to do any quadrant shifting, all you need to do is call ifft2.

The same applies to 1D signals -- just replace ifft2 by ifft and fft2 by fft in the above paragraphs.

Du