Support:Bugs and solutions

How this system works

1. If you encounter an error, please read the 'Known bugs and solutions' section below to find if this error has been reported and solved. If not, please report this bug in the 'Report new bugs' section.
2. As this page is being 'watched' by COMKAT developers, any changes to this page will trigger the wiki to send us an e-mail notifying us of the new bug report.
3. Then we will look at the bug report, provide an answer and solution and reply to the user who reported it via e-mail.
4. Once this bug is resolved, we will move the report from the 'Report new bugs' section to the 'Known bugs and solutions' section.

NOTE: This page will not contain the bug reports for the previous releases of COMKAT if this bug has been resolved in the new version. Meriahkan pesta ulang tahun bersama GarudaFood Amikom.us Tempat Belanja Hosting Murah

How to submit a bug report

1. Search to see if this bug has been reported previously. Use the wiki search functions at the top of this page.
2. Get the following information of your system:
   1. Operating system (and whether it's a 64-bit system with the MATLAB command 'computer')
   2. MATLAB version (use commands 'ver' and 'mexext' in MATLAB)
   3. COMKAT version
3. Copy the following text.

Bug reported by Xin Li. Email: xli51@jhu.edu

OS: Microsoft Windows XP Version 5.1 (Build 2600: Service Pack 3) (32bit)

MATLAB: MATLAB Version 7.1.0.246 (R14) Service Pack 3

COMKAT: R3.2

Bug description: 
When using the 'analyzeFit', there is always error information below. 
??? Error using ==> mtimes
Inner matrix dimensions must agree.
Error in ==> compartmentModel.analyzeFit at 34
if (abs(Jot'*d) > 0.1* abs(d.'*A*d)),

But when I try to run the following code in the 'analyzeFit' function separately, it works.

fjac = full(jac);
[uu,ss,vv]=svd(fjac'*fjac); % jac is to include weighting
covar = vv*diag(1./diag(ss))*uu';
corr = covar./(sqrt(diag(covar)) * sqrt(diag(covar))');
sdev = sqrt(diag(covar));

% rfm2 20021212 added test for gradient
% covariance matrix is calculated assuming gradient is "small".  test this 
d = 0.005*pfit(:) + 1e-10;  % small pertubation to parameter estimate
Jot = fjac'*residual(:);
A = fjac'*fjac;  % approximate Hessian
if (abs(Jot'*d) > 0.1* abs(d.'*A*d)),
    fprintf(['analyzeFit: Covariance matrix estimate is not valid because gradient is ' ...
        'too large.  \nTry fitting again but with more stringent convergence tolerance.\n']);
end


Attachment(optional):

Additional contact information (optional):

4. Then click 'Edit' (in the footer) near the bottom of this page. (You must be logged in to edit.)
5. Paste the text into the edit box under '== New bugs waiting for response =='.
6. Edit the pasted text according to the bug you found.
7. When finished click 'Save page'. You may also include attachments such as .mat or .m files.

Go to Support:Detailed tutorial on bug report for a more detailed step-by-step instruction with pictures.

Known bugs and solutions

Download and installation

Problem: Cannot download COMKAT

When I clicked the link to COMKAT download, the download did not start and I got a 'HTTP 403 Forbidden' error message.

Solution: Log in before you download the COMKAT files.

The COMKAT web server requires you to register and log in before you download the COMKAT files.

Problem: fdgcmd.m in the examples folder crashes with an error

The fitting function in the example fdgcmd.m returns this error

fdgcmd('INVERSE') plotOption = 1 ??? Input argument "cm" is undefined. 
Error in ==> fitfunc at 54 parmToChange = get(cm, 'SensitivityName'); 
Error in ==> optim\private\snls at 423 [newfvec,newA] = feval(funfcn{3},xargin{:},varargin{:}); 
Error in ==> optim\private\lsqncommon at 220 [x,FVAL,LAMBDA,JACOB,EXITFLAG,OUTPUT,msg]=... 
Error in ==> lsqcurvefit at 174 [x,Resnorm,FVAL,EXITFLAG,OUTPUT,LAMBDA,JACOB] = ... 
Error in ==> compartmentModel.fit at 145 [newParmVal,resnorm,residual,exitFlag,output,lambda,jacobian] = lsqcurvefit('fitfunc', initParmVal, [], data, lb, ub, ...
Error in ==> fdgcmd at 208 pfit = fit(cm, pinit, lb, ub);

Solution: 'lsqcurvefit' was changed in MATLAB 2006a

This error arises because the newer MATLAB function

lsqcurvefit

has a different argument passing. In MATLAB version earlier than R2006, the lsqcommon passes an extra argument which is empty. That is why there is an unused argument

fitfunc.m

. However, this is fixed in R2006a and later so now the cm is passed as "unused". If you are using MATLAB 2006a or later, to solve this problem as a workaround, edit the

fitfunc.m

and insert the following lines in line 38: if nargin<3 cm=unused; end

Problem: Parameter estimates obtained using two different computers are different

I use two computers, one xp sp1 with MATLAB v7 sp1, the other is sp2. The estimated value from both computers are different. I have checked many times that I use the same data. Would it be possible to happen?

Solution: This behavior is to be expected

The default optimizer of COMKAT is lsqcurvefit. As it is a function of the MATLAB optimization toolbox, in different versions of MATLAB the fitting results may differ. So the fitted values could be slightly different under different service packs. However, the optimizer

LMWLS

is a function of COMKAT so it will not change in different MATLAB versions. So if you use this optimizer you are more likely to get identical parameter estimates regardless the MATLAB version. On the other hand, there may be platform differences that cause numeric roundoff to be different and this could impact parameter estimates so small differences in estimates are not unexpected.

Problem: Undefined command/function 'pxEval'

I just downloaded and attempted to install the latest version of COMKAT. When attempting to validate the first calculation, I get the following error message:

??? Undefined command/function 'pxEval'.

====S