Nabble - Mathematics

Problems with tests of statistics package

2009-12-21T09:50:22Z

Hi,
I have some issues with the statistics package testing with Octave 3.2.3
and would like to ask whether other people see the same result:

[regress]
warning: inverse: matrix singular to machine precision, rcond = 3.50566e-20

=======================================================================

[linkage]
warning: dmult is obsolete and will be removed from a future version of
Octave; please use the straightforward (and now efficient) syntax
"diag(A)*B".
***** warning <monotonically> linkage (pdist (x), "centroid");
!!!!! expected <monotonically> but got inverse: matrix singular to
machine precision, rcond = 3.50566e-20
shared variables {
x =

0 1 4
3 2 3
1 2 2
3 3 0
0 4 4
4 4 3
1 1 4
2 1 0
4 2 0
3 2 0
0 2 1
1 3 1

t = 1.0000e-06
}

=======================================================================

[__tbl_delim__]
***** test
[d err] = __tbl_delim__ ("bars");
assert (isnan (d));
assert (! isempty (err));
!!!!! test failed
sprintf: wrong type argument `cell'

Thanks
Thomas

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: axis equal

2009-12-20T00:48:51Z

phil1 wrote:
> Are any alternate plotting backends available on Windows?
> available_backends returns only gnuplot. Is that a limitation of the
> Octave-forge installer or Octave on Windows in general?

Yes. Issue
backend("fltk");

However, it's not yet fully implemented, so e.g. text display does not
work. See the news.txt and readme.txt files
A more fully functional fltk backend is currently available only in
the development sources.

benjamin

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

optim/inst/leasqr.m: only compute additional data if needed

2009-12-19T06:41:39Z

Hi,

the attached patch to leasqr again does some optimization. This is the
last of the changes I planned to do for now. I'll commit it soon,
since it can still be discussed and possibly changed afterwards.

Changes:

1. Do not compute additional data (covariance matrix of parameters and
such) if they are not requested.

2. Check whether additional data can be reliably computed and if not,
return matrices of 'NA' of respective dimensions (instead of
triggering warnings and sometimes returning Inf/NaN, sometimes
returning unreliable data).

3. Reject data which lead to non-real residuals since algorithm is not
suitable for these and can not easily be changed to be suitable
(as far as I can see ...).

4. Accept row-vector for argument 'x'.

5. Some code-optimization.

6. Some helptext cleanup.

Regards,

Olaf

Index: leasqr.m =================================================================== --- leasqr.m (revision 6655) +++ leasqr.m (working copy) @@ -25,22 +25,24 @@ %% %% Version 3.beta %% Optional parameters are in braces {}. - %% x = column vector or matrix of independent variables, 1 row per - %% observation: x = [x0 x1....xm]. - %% y = column vector of observed values, same number of rows as x. - %% wt = column vector (dim=length(x)) of statistical weights. These - %% should be set to be proportional to (sqrt of var(y))^-1; (That is, - %% the covariance matrix of the data is assumed to be proportional to + %% x = vector or matrix of independent variables, 1 entry or row per + %% observation. + %% y = vector of observed values, same length as x or as number of + %% rows of x. + %% wt = vector (dim=length(y)) of statistical weights. These should + %% be set to be proportional to (sqrt of var(y))^-1; (That is, the + %% covariance matrix of the data is assumed to be proportional to %% diagonal with diagonal equal to (wt.^2)^-1. The constant of %% proportionality will be estimated.); default = ones(length(y),1). - %% pin = column vec of initial parameters to be adjusted by leasqr. + %% pin = vec of initial parameters to be adjusted by leasqr. %% dp = fractional increment of p for numerical partial derivatives; %% default = .001*ones(size(pin)) %% dp(j) > 0 means central differences on j-th parameter p(j). %% dp(j) < 0 means one-sided differences on j-th parameter p(j). %% dp(j) = 0 holds p(j) fixed i.e. leasqr wont change initial guess: pin(j) - %% F = name of function in quotes; the function shall be of the form y=f(x,p), - %% with y, x, p of the form y, x, pin as described above. + %% F = name of function in quotes or function handle; the function + %% shall be of the form y=f(x,p), with y, x, p of the form y, x, pin + %% as described above; the returned y must be a column vector. %% dFdp = name of partial derivative function in quotes; default is 'dfdp', a %% slow but general partial derivatives function; the function shall be %% of the form prt=dfdp(x,f,p,dp,F[,bounds]). For backwards @@ -88,9 +90,9 @@ %% covr = diag(covariance matrix of the residuals). %% stdresid = standardized residuals. %% Z = matrix that defines confidence region (see comments in the source). - %% r2 = coefficient of multiple determination. + %% r2 = coefficient of multiple determination, intercept form. %% - %% All Zero guesses not acceptable + %% All Zero guesses not acceptable. Not suitable for non-real residuals. %% The following two blocks of comments are chiefly from the original %% version for Matlab. For later changes the logs of the Octave Forge @@ -183,10 +185,11 @@ %% y=y(:); wt=wt(:); pin=pin(:); dp=dp(:); %change all vectors to columns + if (isvector (x)) x = x(:); end %% check data vectors- same length? - m=length(y); n=length(pin); [m1,m2]=size(x); - if (m1~=m) - error('input(x)/output(y) data must have same number of rows ') + m=length(y); n=length(pin); + if (size (x, 1) ~= m) + error('input(x)/output(y) data must have same length ') end %% processing of 'options' @@ -251,9 +254,9 @@ p = pin; f=feval(F,x,p); fbest=f; pbest=p; r=wt.*(y-f); - ss=r'*r; + if (~isreal (r)) error ('weighted residuals are not real'); end + ss = r.' * r; sbest=ss; - nrm=zeros(n,1); chgprev=Inf*ones(n,1); cvg=0; epsLlast=1; @@ -262,30 +265,23 @@ %% do iterations %% for iter=1:niter + nrm = zeros (1, n); pprev=pbest; prt = eval (dfdp_cmd); r=wt.*(y-fbest); + if (~isreal (r)) error ('weighted residuals are not real'); end sprev=sbest; sgoal=(1-stol)*sprev; - for j=1:n - if (dp(j)==0) - nrm(j)=0; - else - prt(:,j)=wt.*prt(:,j); - nrm(j)=prt(:,j)'*prt(:,j); - if (nrm(j)>0) - nrm(j)=1/sqrt(nrm(j)); - end - end - prt(:,j)=nrm(j)*prt(:,j); - end - %% above loop could ? be replaced by: - %% prt=prt.*wt(:,ones(1,n)); - %% nrm=dp./sqrt(diag(prt'*prt)); - %% prt=prt.*nrm(:,ones(1,m))'; + msk = dp ~= 0; + prt(:, msk) = prt(:, msk) .* wt(:, ones (1, sum (msk))); + nrm(msk) = sumsq (prt(:, msk)); + msk = nrm > 0; + nrm(msk) = 1 ./ sqrt (nrm(msk)); + prt = prt .* nrm(ones (1, m), :); + nrm = nrm.'; [prt,s,v]=svd(prt,0); s=diag(s); - g=prt'*r; + g = prt.' * r; for jjj=1:length(epstab), epsL = max(epsLlast*epstab(jjj),1e-7); se=sqrt((s.*s)+epsL); @@ -296,7 +292,7 @@ idx = ~isinf(maxstep); limit = abs(maxstep(idx).*pprev(idx)); chg(idx) = min(max(chg(idx),-limit),limit); - if (verbose & any(ochg ~= chg)) + if (verbose && any(ochg ~= chg)) disp(['Change in parameter(s): ', ... sprintf('%d ',find(ochg ~= chg)), 'were constrained']); end @@ -321,7 +317,8 @@ %% f=feval(F,x,p); r=wt.*(y-f); - ss=r'*r; + if (~isreal (r)) error ('weighted residuals are not real'); end + ss = r.' * r; if (ss<sbest) pbest=p; fbest=f; @@ -341,7 +338,7 @@ end aprec=abs(pprec.*pbest); %% [aprec, chg, chgprev] - if (all(abs(chg) < aprec) & all(abs(chgprev) < aprec)) + if (all(abs(chg) < aprec) && all(abs(chgprev) < aprec)) cvg=1; if (verbose) fprintf('Parameter changes converged to specified precision\n'); @@ -363,6 +360,8 @@ cvg=((sbest>sgoal)|(sbest<=eps)|cvg); if (cvg ~= 1) disp(' CONVERGENCE NOT ACHIEVED! '); end + if (~(verbose || nargout > 4)) return; end + %% CALC VARIANCE COV MATRIX AND CORRELATION MATRIX OF PARAMETERS %% re-evaluate the Jacobian at optimal values jac = eval (dfdp_cmd); @@ -384,58 +383,88 @@ Qinv = diag (tp); end resid=y-f; %un-weighted residuals - tp = resid' * Qinv * resid; + if (~isreal (r)) error ('residuals are not real'); end + tp = resid.' * Qinv * resid; covr = (tp / m) * Q; %covariance of residuals - Qinv = ((m - n) / tp) * Qinv; % Qinv now contains the inverse of the - % guessed covariance matrix of the data. - % No new variable was used, but later - % calculations and comments using Qinv - % remain valid with the new Qinv. - %% argument of inv may be singular - covp = inv (jac' * Qinv * jac); % simplified Eq. 7-5-13, Bard %cov of - % parm est - d=sqrt(diag(covp)); - corp=covp./(d*d'); + %% Matlab compatibility and avoiding recomputation make the following + %% logic clumsy. + compute = 1; + if (m <= n) + compute = 0; + else + Qinv = ((m - n) / tp) * Qinv; + %% simplified Eq. 7-5-13, Bard; cov of parm est, inverse; outer + %% parantheses contain inverse of guessed covariance matrix of data + covpinv = jac.' * Qinv * jac; + if (exist ('rcond') && rcond (covpinv) <= eps) + compute = 0; + elseif (rank (covpinv) < n) + %% above test is not equivalent to 'rcond' and may unnecessarily + %% reject some matrices + compute = 0; + end + end + if (compute) + covp = inv (covpinv); + d=sqrt(diag(covp)); + corp = covp ./ (d * d.'); + else + covp = NA * ones (n); + corp = covp; + end if (exist('sparse')) covr=spdiags(covr,0); - stdresid=resid ./ sqrt (covr); else covr=diag(covr); % convert returned values to % compact storage - stdresid=resid ./ sqrt (covr); end - Z=((m-n)*jac'*Qinv*jac)/(n*resid'*Qinv*resid); + stdresid = resid .* abs (wt) / sqrt (tp / m); % equivalent to resid ./ + % sqrt (covr) + if (~(verbose || nargout > 8)) return; end + + if (m > n) + Z = ((m - n) / (n * resid.' * Qinv * resid)) * covpinv; + else + Z = NA * ones (n); + end + %%% alt. est. of cov. mat. of parm.:(Delforge, Circulation, 82:1494-1504, 1990 %%disp('Alternate estimate of cov. of param. est.') %%acovp=resid'*Qinv*resid/(m-n)*inv(jac'*Qinv*jac); - %%Calculate R^2 (Ref Draper & Smith p.46) + if (~(verbose || nargout > 9)) return; end + + %%Calculate R^2, intercept form %% - r=corrcoef([y(:),f(:)]); - r2=r(1,2).^2; + tp = sumsq (y - mean (y)); + if (tp > 0) + r2 = 1 - sumsq (resid) / tp; + else + r2 = NA; + end %% if someone has asked for it, let them have it %% if (verbose) eval(plotcmd); disp(' Least Squares Estimates of Parameters') - disp(p') + disp(p.') disp(' Correlation matrix of parameters estimated') disp(corp) disp(' Covariance matrix of Residuals' ) disp(covr) disp(' Correlation Coefficient R^2') disp(r2) - sprintf(' 95%% conf region: F(0.05)(%.0f,%.0f)>= delta_pvec''*Z*delta_pvec',n,m-n) + sprintf(' 95%% conf region: F(0.05)(%.0f,%.0f)>= delta_pvec.''*Z*delta_pvec',n,m-n) Z %% runs test according to Bard. p 201. - n1 = sum((f-y) < 0); - n2 = sum((f-y) > 0); - nrun=sum(abs(diff((f-y)<0)))+1; - if ((n1>10)&(n2>10)) % sufficent data for test? + n1 = sum (resid > 0); + n2 = sum (resid < 0); + nrun=sum(abs(diff(resid > 0)))+1; + if ((n1 > 10) && (n2 > 10)) % sufficent data for test? zed=(nrun-(2*n1*n2/(n1+n2)+1)+0.5)/(2*n1*n2*(2*n1*n2-n1-n2)... /((n1+n2)^2*(n1+n2-1))); if (zed < 0)

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: optim/inst/leasqr.m: correction of covariance matrices and standard residuals

2009-12-18T06:14:01Z

Olaf,

I am sorry I was not able to spend time testing your patches. But
a glance over them does not show anything wrong especially since your
spotted complex conjugate issue your self. I will be quite busy for the
rest of the weak. Then I will be able to assign a time slot for testing,
but in any case I would prefer to work with already modified SVN version.
So we would have a common ground for patches exchange.

On Fri, Dec 18, 2009 at 10:18:04AM +0100, Olaf Till wrote:

> On Thu, Dec 17, 2009 at 01:41:21PM +0100, Olaf Till wrote:
> > On Wed, Dec 16, 2009 at 01:54:57PM +0100, Olaf Till wrote:
> > > Hi,
> > >
> > > the attached patch does some corrections and code-optimization to
> > > optim/inst/leasqr.m.
> >
> > It was probably a mistake to use the transpose instead of the
> > conjugate transpose in some places in the submitted patch. So I'll use
> > the conjugate transpose, as in the original code, at least until I
> > have looked into the matter of complex data more thoroughly.
>
> If there are no comments and nobody says he needs more time for
> commenting, I would take this as a sign that I can commit today.

--
Eugeniy

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Problem with MEANDEV()

2009-12-18T04:49:04Z

On Fri, Dec 18, 2009 at 1:38 PM, Ben Abbott <bpabbott@...> wrote:

>
> On Dec 18, 2009, at 1:40 AM, Bruce Axtens wrote:
>
>> I'm getting a weird answer from meandev as compared to Excel, Wolfram Alpha and Protium.
>>
>> GNU Octave, version 3.2.3
>> Octave was configured for "i686-pc-mingw32"
>>
>> meandev([8, 8, 9, 11, 11, 15, 16, 19, 19, 20, 20, 21, 21, 25, 32]) --> ans = 5.8571
>>
>> All the others give 5.46667
>> e.g. WolframAlpha
>> http://www.wolframalpha.com/input/?i=mean+deviation%288%2C+8%2C++9%2C+11%2C+11%2C+15%2C+16%2C+19%2C+19%2C+20%2C+20%2C+21%2C+21%2C+25%2C+32%29
>>
>> What am I doing wrong?
>>
>> Kind regards,
>> Bruce.
>
> Looking at the link below ...
>
> http://mathworld.wolfram.com/MeanDeviation.html
>
> ... the mean deviation may be calculated as ...
>
> meandev = @(x) mean (abs(x-mean(x)));
> meandev([8, 8, 9, 11, 11, 15, 16, 19, 19, 20, 20, 21, 21, 25, 32])
> ans = 5.4667
>
> So it does appear that you have found a bug.
>
> Unfortunately, you've reported this to the wrong mail list. help-octave@... is for Octave and its core functions. The function, meandev, is part of the statistics package at Octave-Forge. The correct mail list for that is octave-dev@....
>
> I've removed help-octave@... and added octave-dev@... to the recipients.
>
> Ben
>
>
>

It's not a bug, it's that meandev divides by n-1 rather than n.
Maximum likelihood estimation calls for dividing by n. There is an
option for unbiased estimation, too, which calls for dividing by n-1.

octave:1> 5.46667*15/14
ans = 5.8571
octave:2>

Cheers, Michael

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Problem with MEANDEV()

2009-12-18T04:38:03Z

On Dec 18, 2009, at 1:40 AM, Bruce Axtens wrote:

> I'm getting a weird answer from meandev as compared to Excel, Wolfram Alpha and Protium.
>
> GNU Octave, version 3.2.3
> Octave was configured for "i686-pc-mingw32"
>
> meandev([8, 8, 9, 11, 11, 15, 16, 19, 19, 20, 20, 21, 21, 25, 32]) --> ans = 5.8571
>
> All the others give 5.46667
> e.g. WolframAlpha
> http://www.wolframalpha.com/input/?i=mean+deviation%288%2C+8%2C++9%2C+11%2C+11%2C+15%2C+16%2C+19%2C+19%2C+20%2C+20%2C+21%2C+21%2C+25%2C+32%29
>
> What am I doing wrong?
>
> Kind regards,
> Bruce.

Looking at the link below ...

http://mathworld.wolfram.com/MeanDeviation.html

... the mean deviation may be calculated as ...

meandev = @(x) mean (abs(x-mean(x)));
meandev([8, 8, 9, 11, 11, 15, 16, 19, 19, 20, 20, 21, 21, 25, 32])
ans = 5.4667

So it does appear that you have found a bug.

Unfortunately, you've reported this to the wrong mail list. help-octave@... is for Octave and its core functions. The function, meandev, is part of the statistics package at Octave-Forge. The correct mail list for that is octave-dev@....

I've removed help-octave@... and added octave-dev@... to the recipients.

Ben

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Maximally flat (maxflat) digital filter design

2009-12-18T01:56:35Z

Hello-

I have written a replacement for the Matlab Signal Processing Toolbox
function "maxflat", which designs maximally flat digital filters of
arbitrary order. Please see
http://sourceforge.net/projects/maxflat-octmat/ for the code and
details. I would be pleased to contribute it to the Octave-Forge Signal
package if there is any interest.

Regards,
Michael

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: optim/inst/leasqr.m: correction of covariance matrices and standard residuals

2009-12-18T01:18:04Z

On Thu, Dec 17, 2009 at 01:41:21PM +0100, Olaf Till wrote:

> On Wed, Dec 16, 2009 at 01:54:57PM +0100, Olaf Till wrote:
> > Hi,
> >
> > the attached patch does some corrections and code-optimization to
> > optim/inst/leasqr.m.
>
> It was probably a mistake to use the transpose instead of the
> conjugate transpose in some places in the submitted patch. So I'll use
> the conjugate transpose, as in the original code, at least until I
> have looked into the matter of complex data more thoroughly.

If there are no comments and nobody says he needs more time for
commenting, I would take this as a sign that I can commit today.

Olaf

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: axis equal

2009-12-17T10:15:05Z

Are any alternate plotting backends available on Windows?
available_backends returns only gnuplot. Is that a limitation of the Octave-forge installer or Octave on Windows in general?

I would try the fltk backend on Ubuntu but I only have Jaunty installed which only has Octave 3.0.x

Phil

Robert T. Short wrote:

This is probably a gnuplot problem.

Phil Fong wrote:
> I am using Octave 3.2.3 on Windows installed with the Octave-forge installer. "axis equal" does not equalize the data aspect ratio of plots. It gets closer but if I plot a square it is still not square.

From forum: octave-dev

Re: optim/inst/leasqr.m: correction of covariance matrices and standard residuals

2009-12-17T04:41:21Z

On Wed, Dec 16, 2009 at 01:54:57PM +0100, Olaf Till wrote:
> Hi,
>
> the attached patch does some corrections and code-optimization to
> optim/inst/leasqr.m.

It was probably a mistake to use the transpose instead of the
conjugate transpose in some places in the submitted patch. So I'll use
the conjugate transpose, as in the original code, at least until I
have looked into the matter of complex data more thoroughly.

Olaf

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: playrec not part of windows distribution

2009-12-17T00:52:44Z

Hi!

2009/12/16 Benjamin Lindner <lindnerben@...>:
> Would you be willing to test if and how playrec can be compiled
> with the mingw compiler shipped with octave/mingw32?

Sure!
Unfortunately I am everything but a compiler-expert, but I see what I can do...

Cheers,
Rene

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: axis equal

2009-12-16T17:47:40Z

This is probably a gnuplot problem.

Phil Fong wrote:

> I am using Octave 3.2.3 on Windows installed with the Octave-forge installer. "axis equal" does not equalize the data aspect ratio of plots. It gets closer but if I plot a square it is still not square.
>
> I see a thread in March that included a patch against 3.1.54 about the behavior of "axis equal" and "axis square". Is that included in 3.2.3? Looking at axis.m and the patch, it seems like it's there.
>
> Phil
>
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.Net email is sponsored by the Verizon Developer Community
> Take advantage of Verizon's best-in-class app development support
> A streamlined, 14 day to market process makes app distribution fast and easy
> Join now and get one step closer to millions of Verizon customers
> http://p.sf.net/sfu/verizon-dev2dev
> _______________________________________________
> Octave-dev mailing list
> Octave-dev@...
> https://lists.sourceforge.net/lists/listinfo/octave-dev
>
>
>

From forum: octave-dev

axis equal

2009-12-16T16:59:36Z

I am using Octave 3.2.3 on Windows installed with the Octave-forge installer. "axis equal" does not equalize the data aspect ratio of plots. It gets closer but if I plot a square it is still not square.

I see a thread in March that included a patch against 3.1.54 about the behavior of "axis equal" and "axis square". Is that included in 3.2.3? Looking at axis.m and the patch, it seems like it's there.

Phil

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: MarkerSize has no effect

2009-12-16T16:47:49Z

>
> plot (1:10,"o", "markersize",5)
> works without any porblem on Octave 3.2.3-2_i686-pc-mingw32_gcc-4.4.0
>
> Octave is case sensitive.
> Please do not use 'MarkerSize' but use 'markaersize' as described in the manual.
>

Both "MarkerSize" and "markersize" seem to have the desired effect on the marker size for markers other than ".".

Phil

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: MarkerSize has no effect

2009-12-16T15:42:35Z

Phil Fong wrote:
> It does not seem possible to draw filled in circle markers of different sizes. "plot(0:10,0:2:20,'o','MarkerSize',10, 'MarkerFaceColor', 'b')" gives circles with a dot in the center. But that seems to be due to MarkerFaceColor having no effect. "plot(0:10,0:2:20,'s','MarkerSize',10, 'MarkerFaceColor', 'r')" produces filled in blue squares instead of squares with blue outlines and red fill.
>
See the recent thread

http://old.nabble.com/make-%22plot%22-draw-filled-dots-to26796661.html

and the older thread

http://old.nabble.com/Plot-Bug%3A-%27markerfacecolor%27-does-not-work-tt25562839.html

it refers to

D.

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: MarkerSize has no effect

2009-12-16T15:09:11Z

Hello

plot (1:10,"o", "markersize",5)
works without any porblem on Octave 3.2.3-2_i686-pc-mingw32_gcc-4.4.0

Octave is case sensitive.
Please do not use 'MarkerSize' but use 'markaersize' as described in the manual.

Regards

Tatsuro

--- Phil Fong <fongpwf@...> wrote:

> I am using Octave 3.2.3-2_i686-pc-mingw32_gcc-4.4.0 installed using the Octave-forge
> installer. As far as I can tell setting MarkerSize for point (.) markers has no effect.
> "plot(0:10,0:2:20,'.','MarkerSize',10)" and "plot(0:10,0:2:20,'.','MarkerSize',10)" produce the
> same plot.
>
> Is this intentional? Matlab does scale point markers with MarkerSize.
>
> It does not seem possible to draw filled in circle markers of different sizes.
> "plot(0:10,0:2:20,'o','MarkerSize',10, 'MarkerFaceColor', 'b')" gives circles with a dot in the
> center. But that seems to be due to MarkerFaceColor having no effect.
> "plot(0:10,0:2:20,'s','MarkerSize',10, 'MarkerFaceColor', 'r')" produces filled in blue squares
> instead of squares with blue outlines and red fill.
>
> Phil
>
>
>
>
>
> ------------------------------------------------------------------------------
> This SF.Net email is sponsored by the Verizon Developer Community
> Take advantage of Verizon's best-in-class app development support
> A streamlined, 14 day to market process makes app distribution fast and easy
> Join now and get one step closer to millions of Verizon customers
> http://p.sf.net/sfu/verizon-dev2dev
> _______________________________________________
> Octave-dev mailing list
> Octave-dev@...
> https://lists.sourceforge.net/lists/listinfo/octave-dev
>

--------------------------------------
Get Disney character's mail address on Yahoo! Mail
http://pr.mail.yahoo.co.jp/disney/

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T13:42:53Z

>> * Search doesn't work if you don't know the shop-talk.
>
> I would claim that nothing works if you don't know the shop-talk.

Well put :)

>
> I must, however, stress that I don't have the time to actually implement
> this, so somebody else would have to do the leg-work...

Admittedly, I also would not have the time to do this.
Using mail clients is quick, nicely customizable to my personal
likes/dislikes, to filtering, marking, tagging, and it requires
next to none administrative work.
And as spare time is one of the thing I believe many do not have
abundant...

benjamin

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T13:36:51Z

> * You don't need to register.

But to post I obviously need to register, so it's not easier here.

> You can go there, search for a solution
> to your problem. In most cases that should be enough.

This is also possible on the mailing list archives, e.g. nabble.

And I miss the threaded tree view on the aforementioned exemplary forum.

benjamin

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T13:27:54Z

Jaroslav Hajek wrote:

> You see, most of the convenience of working with a mailing list comes
> from the capabilities of existing mail clients. Of course, if you read
> mail in plain text from command line, any kind of web forum must look
> good, but for instance GMail is quite smart and customizable, and I
> don't think it's that easy to outmatch.

Plus, using mail clients, one can do work offline. That's a feature
I like because I don't need to be online when reading/searching.
It's in a way like distributed VCS.

benjamin

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: playrec not part of windows distribution

2009-12-16T13:13:03Z

Rene Teerwill wrote:
> Hello!
>
> The playrec-function is not part of the windows-installer even though
> this is claimed on various download sites e.g. on
> http://www.softpedia.com/get/Science-CAD/Octave.shtml

I believe it has been mentioned before, but here it goes again:
What somebody on some website completely unrelated to octave may or may
not claim with respect to what's or what's not inluded in octave is -
frankly - irrelevant.

> In the meantime I've learned it has been part of octave in version 3.0.3
> Is there a plan to add it to the distribution again?

I took a look at http://www.playrec.co.uk/ and from the documentation
there I gather they require both msvc and SDK headers to compile playrec.

To be included in a binary distribution, the source code has to be
compatible to a GNU license and compilable with gnu gcc.
There are legal issues with distributing binaries compiled with msvc
under a GNU license. This has been already been discussed on the
mailing list, so I won't elaborate on this again.

So requiring msvc means: no, it will not be bundled in the distribution.

I'm not sure on how the SDK headers fit into the GNU license.

Would you be willing to test if and how playrec can be compiled
with the mingw compiler shipped with octave/mingw32?

benjamin

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T13:11:53Z

On Wed, Dec 16, 2009 at 2:44 PM, Søren Hauberg <soren@...> wrote:
> ons, 16 12 2009 kl. 14:16 -0500, skrev Judd Storrs:
>> Just playing devils advocate here for a bit.
>
> :-) Then I'll ask some stupid questions (because I actually don't know
> the answer...)

Sure, however, I will say that I believe the wiki is far superior to a forum.

>> * Sticky posts i.e. dedicated threads that hang around. For a while
>> there was a lot of trouble compiling octave on ubuntu and it really
>> wasn't that easy to find the old posts.
>
> Who would mark threads as sticky? (and unmark the threads later on)

In forums I think that's done by section moderators. I've seen forums
where the sub-categories are set up with some accounts able to
moderate within specific sections. In some forums each section sort of
develops it's own sort of community but it really takes dedication for
that to work out. In any case, I prefer wiki pages over forums.

>> * Forum sections would be nice and search doesn't always work out.
>> For example if I want to know what's going on in image processing in
>> octave, I'm not sure all posts will have the term "image processing".
>
> Who would determine what goes in which threads?

In forums this is based on where the submitter posted the question.
i.e. the submitter asking about an image processing question would
post the question to the "Image Processing" section of the forum. I
don't know how that could work for things that originate from a
mailing list backend...

>> * Forums can be easier to browse if you're just curious and want to
>> learn. For example there could be a forum on vectorization tricks. How
>> would you search for that?
>
> Often people start out with asking a question about one thing. The
> thread then slowly changes topic and some neat trick about vectorisation
> is given. Who would make sure this trick would end up in a vectorisation
> section?

That is definitely true. I think the forum approach would rely on the
section structure and expect questions to be posted inside a section
perhaps called "Performance optimization".

>> * Subscribing to a mailing list does seem to require a greater level
>> of commitment than a forum including setting up filtering rules etc.
>> Occasionally, I've hit a forum and I can answer a question I'll just
>> create a quick account and join the discussion.
>
> Good point, although I must admit I'm not sure it's actually "quick" to
> create an account.

It depends. OpenID could reduce that hurdle if someone wanted to go that route.

>> * Search doesn't work if you don't know the shop-talk.
>
> I would claim that nothing works if you don't know the shop-talk. How
> can you find the 'vectorisation' section if you don't know said term?

I think the idea is that forum sections can act like an index. Sure
you may not know that vectorization is maybe what you want but if you
see a section called "Performance" with perhaps a subsection called
"Vectorization" you can click around and discover some things. Of
course you can also click around and discover things on a wiki. The
wiki is a dramatically superior approach for presenting learnable
material, IMHO.

>> Presumably an alternative would be to create separate mailing lists
>> for each octave-help-ubuntu@, octave-help-image-processing@,
>> octave-help-vectorization-tricks@, ... But to me that seems even worse
>> than doing nothing.
>
> Agreed. That would be worse.

R has setup some "special interest group" mailing lists. I don't think
we have nearly as much volume as R.

> I must, however, stress that I don't have the time to actually implement
> this, so somebody else would have to do the leg-work...

I think my feelings are aligned with your's and Jaroslav's on this. If
the forum was there, honestly I'm not sure I would use. If all of us
are using the mail backend, what's the point? The huge down-side of
running a forum is that it does require dedication and vigilance both
on the administration and moderation fronts. If enough people are
interested, active moderation work can be handled by deputies. If not,
then it's a burden. Honestly, help@ hasn't been very high volume since
I've joined, anyway. I personally don't see the effort/benefit curve
working out at the moment.

--judd

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

MarkerSize has no effect

2009-12-16T12:07:47Z

I am using Octave 3.2.3-2_i686-pc-mingw32_gcc-4.4.0 installed using the Octave-forge installer. As far as I can tell setting MarkerSize for point (.) markers has no effect. "plot(0:10,0:2:20,'.','MarkerSize',10)" and "plot(0:10,0:2:20,'.','MarkerSize',10)" produce the same plot.

Is this intentional? Matlab does scale point markers with MarkerSize.

It does not seem possible to draw filled in circle markers of different sizes. "plot(0:10,0:2:20,'o','MarkerSize',10, 'MarkerFaceColor', 'b')" gives circles with a dot in the center. But that seems to be due to MarkerFaceColor having no effect. "plot(0:10,0:2:20,'s','MarkerSize',10, 'MarkerFaceColor', 'r')" produces filled in blue squares instead of squares with blue outlines and red fill.

Phil

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: I want to contribute to octave project.

2009-12-16T11:50:25Z

Hi

ons, 16 12 2009 kl. 23:08 +0800, skrev 陈锟:
> Could I begin with The ocatve manual?I think it's the
> important thing for common users.

Please keep your replies on the list (press "reply to all" instead of
"reply") such that others can follow and reply.

I don't know which place is the best to start. The Octave manual
consists of text _and_ help texts from the individual functions. So, if
you translate the manual you also end up translating the individual
functions.

As I previously said, I have no experience with translations, so I don't
know which place is the best to start. Does anybody else have advice
here?

Søren

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T11:44:39Z

ons, 16 12 2009 kl. 14:16 -0500, skrev Judd Storrs:
> Just playing devils advocate here for a bit.

:-) Then I'll ask some stupid questions (because I actually don't know
the answer...)

> * Sticky posts i.e. dedicated threads that hang around. For a while
> there was a lot of trouble compiling octave on ubuntu and it really
> wasn't that easy to find the old posts.

Who would mark threads as sticky? (and unmark the threads later on)

> * Forum sections would be nice and search doesn't always work out.
> For example if I want to know what's going on in image processing in
> octave, I'm not sure all posts will have the term "image processing".

Who would determine what goes in which threads?

> * Forums can be easier to browse if you're just curious and want to
> learn. For example there could be a forum on vectorization tricks. How
> would you search for that?

Often people start out with asking a question about one thing. The
thread then slowly changes topic and some neat trick about vectorisation
is given. Who would make sure this trick would end up in a vectorisation
section?

> * Subscribing to a mailing list does seem to require a greater level
> of commitment than a forum including setting up filtering rules etc.
> Occasionally, I've hit a forum and I can answer a question I'll just
> create a quick account and join the discussion.

Good point, although I must admit I'm not sure it's actually "quick" to
create an account.

> * Search doesn't work if you don't know the shop-talk.

I would claim that nothing works if you don't know the shop-talk. How
can you find the 'vectorisation' section if you don't know said term?

> Presumably an alternative would be to create separate mailing lists
> for each octave-help-ubuntu@, octave-help-image-processing@,
> octave-help-vectorization-tricks@, ... But to me that seems even worse
> than doing nothing.

Agreed. That would be worse.

> Supposedly, ubuntuforums.org is a vBullitin
> frontend to mailman. Maybe that could help?

I agree with Jaroslav who said:

"Of course, if you can transform the list to a web forum while
preserving the mailing list interface (i.e. so that I can read
and
post through e-mail as before), I can't have reasonable
objections."

I must, however, stress that I don't have the time to actually implement
this, so somebody else would have to do the leg-work...

Søren

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T11:16:44Z

Just playing devils advocate here for a bit. Here are some features
that I like when I hit a forum that seem to be missing from
nabble/mailing lists with search.

* Sticky posts i.e. dedicated threads that hang around. For a while
there was a lot of trouble compiling octave on ubuntu and it really
wasn't that easy to find the old posts.

* Forum sections would be nice and search doesn't always work out.
For example if I want to know what's going on in image processing in
octave, I'm not sure all posts will have the term "image processing".

* Forums can be easier to browse if you're just curious and want to
learn. For example there could be a forum on vectorization tricks. How
would you search for that?

* Subscribing to a mailing list does seem to require a greater level
of commitment than a forum including setting up filtering rules etc.
Occasionally, I've hit a forum and I can answer a question I'll just
create a quick account and join the discussion.

* Search doesn't work if you don't know the shop-talk.

Presumably an alternative would be to create separate mailing lists
for each octave-help-ubuntu@, octave-help-image-processing@,
octave-help-vectorization-tricks@, ... But to me that seems even worse
than doing nothing. Supposedly, ubuntuforums.org is a vBullitin
frontend to mailman. Maybe that could help?

--judd

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T10:57:21Z

Hello all,

I just wanted to point out that, as Jaroslav mentioned, there already
exists a forum-like archive of the Octave mailing lists, courtesy of
nabble.com [1]. It's just as good, if not better than other actual
forum software out there -- including search; tree views of threads;
replying to messages from the web interface (and even get alerts);
etc. It has a spartan interface (which IMHO is best), but it
essentially provides what a forum would, and then some.

For example, this thread can be found here:
<http://old.nabble.com/forum/ViewPost.jtp?post=26794685&framed=y>

Perhaps this is an appropriate compromise?

[1] <http://old.nabble.com/Octave-f1895.html>

--
Mansour Moufid

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

optim/inst/leasqr.m: correction of covariance matrices and standard residuals

2009-12-16T04:54:57Z

Hi,

the attached patch does some corrections and code-optimization to
optim/inst/leasqr.m.

Except minor changes to the comment blocks, all changes apply to the
code after the actual parameter estimation.

Comments on changes which alter the output:

To get the covariance matrices of the residuals and of the data, a
certain matrix must be divided by the number of residuals "m" or the
number of degrees of freedom "m - n", respectively. The previous code
erroneously divided by "m - n" for the cov.-matr. of residuals "covr",
and nevertheless calculated the cov.-matr. of data "Vy" from the
latter by multiplying with "m / (m - n)" (reduced to "1/(1-n/m)")), so
"Vy" was also not correct. The two covariance matrices are now
calculated correctly, although of the latter only the inverse is
computed due to some code-optimization. Affected output: "covr",
"covp", "corp".

Although I found almost no reference for this, I think it is more
sensible to normalize the standard residuals "stdresid" against the
square root of the variance of the _residuals_, not the sqare root of
the variance of the _data_. This has been changed accordingly. The
former normalization was also wrong because it used the wrongly
calculated "Vy" (see above). Affected output: "stdresid"

all changes:

- corrected calculation of covariance matrix of residuals and
covariance matrix of data, thereby correcting covariance matrix of
parameters and correlation matrix of parameters

- simplified calculation of covariance matrix of parameters

- standard residuals now normalized against square root of variance of
residuals, not of data

- some repetitive computation of the same data avoided in calculation
of covariance matrix of residuals and later, one repetitive
multiplication of a matrix with a scalar avoided

- removed abs() on non-negative data in line 376

- changed hint to article of Delforge to apply to the currently
existing variables

- helptext: added comment on bounds

Except code-optimization, the above numeric changes can be
equivalently achieved simply with:

--- leasqr.m 2009-12-16 13:13:57.000000000 +0100
+++ tp_leasqr.m 2009-12-16 13:23:09.000000000 +0100
@@ -375,7 +375,7 @@
Qinv=diag(wt.*wt);
end
resid=y-f; %un-weighted residuals
- covr=resid'*Qinv*resid*Q/(m-n); %covariance of residuals
+ covr=resid'*Qinv*resid*Q/m; %covariance of residuals
Vy=1/(1-n/m)*covr; % Eq. 7-13-22, Bard %covariance of the data

jtgjinv=inv(jac'*Qinv*jac); %argument of inv may be
@@ -387,11 +387,11 @@

if (exist('sparse'))
covr=spdiags(covr,0);
- stdresid=resid./sqrt(spdiags(Vy,0));
+ stdresid=resid./sqrt(covr);
else
covr=diag(covr); % convert returned values to
% compact storage
- stdresid=resid./sqrt(diag(Vy)); % compute then convert for compact storage
+ stdresid=resid./sqrt(covr); % compute then convert for compact storage
end
Z=((m-n)*jac'*Qinv*jac)/(n*resid'*Qinv*resid);

You could compare this with the full patch to test whether the same
results are returned (they are, up to 12 digits for "covp" in my
test).

Regards, Olaf

Index: leasqr.m =================================================================== --- leasqr.m (revision 6652) +++ leasqr.m (working copy) @@ -74,6 +74,9 @@ %% for each parameter. Default: [-Inf, Inf] in each row. If this %% field is used with an existing user-side function for 'dFdp' %% (see above) the functions interface might have to be changed. + %% _Warning_: If bounds are set, returned guesses of corp, covp, and + %% Z are generally invalid, even if final parameters are not at the + %% bounds. %% %% OUTPUT VARIABLES %% f = column vector of values computed: f = F(x,p). @@ -89,6 +92,10 @@ %% %% All Zero guesses not acceptable + %% The following two blocks of comments are chiefly from the original + %% version for Matlab. For later changes the logs of the Octave Forge + %% svn repository should also be consulted. + %% A modified version of Levenberg-Marquardt %% Non-Linear Regression program previously submitted by R.Schrager. %% This version corrects an error in that version and also provides @@ -120,6 +127,7 @@ %% 5) revise estimates of corp, stdev %% Modified by Ray Muzic 11-Oct-1992 %% 1) revise estimate of Vy. remove chi2, add Z as return values + %% (later remark: the current code contains no variable Vy) %% Modified by Ray Muzic 7-Jan-1994 %% 1) Replace ones(x) with a construct that is compatible with versions %% newer and older than v 4.1. @@ -367,37 +375,42 @@ %% diag(1/wt.^2). %% cov matrix of data est. from Bard Eq. 7-5-13, and Row 1 Table 5.1 + tp = wt.^2; if (exist('sparse')) % save memory - Q=sparse(1:m,1:m,1./wt.^2); - Qinv=sparse(1:m,1:m,wt.^2); + Q = sparse (1:m, 1:m, 1 ./ tp); + Qinv = sparse (1:m, 1:m, tp); else - Q=diag((0*wt+1)./(wt.^2)); - Qinv=diag(wt.*wt); + Q = diag (ones (m, 1) ./ tp); + Qinv = diag (tp); end resid=y-f; %un-weighted residuals - covr=resid'*Qinv*resid*Q/(m-n); %covariance of residuals - Vy=1/(1-n/m)*covr; % Eq. 7-13-22, Bard %covariance of the data + tp = resid.' * Qinv * resid; + covr = (tp / m) * Q; %covariance of residuals - jtgjinv=inv(jac'*Qinv*jac); %argument of inv may be - %singular - covp=jtgjinv*jac'*Qinv*Vy*Qinv*jac*jtgjinv; % Eq. 7-5-13, Bard %cov of + Qinv = ((m - n) / tp) * Qinv; % Qinv now contains the inverse of the + % guessed covariance matrix of the data. + % No new variable was used, but later + % calculations and comments using Qinv + % remain valid with the new Qinv. + %% argument of inv may be singular + covp = inv (jac.' * Qinv * jac); % simplified Eq. 7-5-13, Bard %cov of % parm est - d=sqrt(abs(diag(covp))); + d=sqrt(diag(covp)); corp=covp./(d*d'); if (exist('sparse')) covr=spdiags(covr,0); - stdresid=resid./sqrt(spdiags(Vy,0)); + stdresid=resid ./ sqrt (covr); else covr=diag(covr); % convert returned values to % compact storage - stdresid=resid./sqrt(diag(Vy)); % compute then convert for compact storage + stdresid=resid ./ sqrt (covr); end Z=((m-n)*jac'*Qinv*jac)/(n*resid'*Qinv*resid); %%% alt. est. of cov. mat. of parm.:(Delforge, Circulation, 82:1494-1504, 1990 %%disp('Alternate estimate of cov. of param. est.') - %%acovp=resid'*Qinv*resid/(m-n)*jtgjinv + %%acovp=resid'*Qinv*resid/(m-n)*inv(jac'*Qinv*jac); %%Calculate R^2 (Ref Draper & Smith p.46) %%

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T04:34:30Z

On Wed, Dec 16, 2009 at 12:59 PM, Ralf Steinhaeusser <kiteralf@...> wrote:
> 2009/12/16 Jaroslav Hajek <highegg@...>:
>> Nice. But besides the search capabilities, I see no big advantages.
>
> Just the main 3:
>
> * You don't need to register. You can go there, search for a solution
> to your problem. In most cases that should be enough.

It's the same with the nabble archive. Anyway, *I* need to register.

> * Privacy: In a forum I don't spread my email address. In a mailing
> list it's very easy to collect email adresses.

How do you mean? The nabble archive doesn't give up the addresses.
They should only be visible to subscribers.

> I've already had to
> abandon 2 of my former email-adresses because they were spammed so
> massively after posting to newsgroups.

I get spam too, yet the spam filter handles them well. just a few
spams per week get through.

> * Only as many emails as I want that.
>
> Again: You can subscribe to the complete forum, then you will get
> everything as email.

That sounds good. But can you also reply to messages via e-mail?

> I personally don't want that.
> I am more or less just interested in the problems I have.
> So I go to the forum, check if there's already a solution
> If there is one I might want to subscribe to it so I get informed when
> there's discussion on that topic again.
> I also subscribe to topics I find to be important, so I get updated
> when there's news and I can find them later easily (there's an area
> "your subscriptions")
> If I see questions where I can help, I post my comments
> Then I post my question and when someone responses I get an email
> automatically (If I've set up my profile that way)
>
> However, I understand that you like to do it differently and I respect
> that. I just thought a different means of communication could make
> life a bit easier, especially for new Octave-users.
>

I understand, but as Soren said, the developers are the most regular
users, so their opinion matters the most.
Of course, if you can transform the list to a web forum while
preserving the mailing list interface (i.e. so that I can read and
post through e-mail as before), I can't have reasonable objections.

--
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T04:01:09Z

uups, my collegue was logged in on my computer - wrong user profile. Sorry.

2009/12/16 Jaroslav Hajek <highegg@...>:
> Nice. But besides the search capabilities, I see no big advantages.

Just the main 3:

* You don't need to register. You can go there, search for a solution
to your problem. In most cases that should be enough.
* Privacy: In a forum I don't spread my email address. In a mailing
list it's very easy to collect email adresses. I've already had to
abandon 2 of my former email-adresses because they were spammed so
massively after posting to newsgroups.
* Only as many emails as I want that.

Again: You can subscribe to the complete forum, then you will get
everything as email.
I personally don't want that.
I am more or less just interested in the problems I have.
So I go to the forum, check if there's already a solution
If there is one I might want to subscribe to it so I get informed when
there's discussion on that topic again.
I also subscribe to topics I find to be important, so I get updated
when there's news and I can find them later easily (there's an area
"your subscriptions")
If I see questions where I can help, I post my comments
Then I post my question and when someone responses I get an email
automatically (If I've set up my profile that way)

However, I understand that you like to do it differently and I respect
that. I just thought a different means of communication could make
life a bit easier, especially for new Octave-users.

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T03:59:01Z

2009/12/16 Jaroslav Hajek <highegg@...>:
> Nice. But besides the search capabilities, I see no big advantages.

Just the main 3:

* You don't need to register. You can go there, search for a solution
to your problem. In most cases that should be enough.
* Privacy: In a forum I don't spread my email address. In a mailing
list it's very easy to collect email adresses. I've already had to
abandon 2 of my former email-adresses because they were spammed so
massively after posting to newsgroups.
* Only as many emails as I want that.

Again: You can subscribe to the complete forum, then you will get
everything as email.
I personally don't want that.
I am more or less just interested in the problems I have.
So I go to the forum, check if there's already a solution
If there is one I might want to subscribe to it so I get informed when
there's discussion on that topic again.
I also subscribe to topics I find to be important, so I get updated
when there's news and I can find them later easily (there's an area
"your subscriptions")
If I see questions where I can help, I post my comments
Then I post my question and when someone responses I get an email
automatically (If I've set up my profile that way)

However, I understand that you like to do it differently and I respect
that. I just thought a different means of communication could make
life a bit easier, especially for new Octave-users.

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: optim/leasqr.m style changes, diff

2009-12-16T03:44:03Z

ons, 16 12 2009 kl. 12:15 +0100, skrev Olaf Till:

> On Tue, Dec 15, 2009 at 09:39:13PM +0100, Søren Hauberg wrote:
> > man, 14 12 2009 kl. 15:17 +0100, skrev Olaf Till:
> > > Anyway I'll keep Matlab compatibility now, but still would like to
> > > make the following style changes beforehand:
> >
> > I think these changes sound just fine (I'm not the maintainer of the
> > package, so my opinion isn't all that important).
>
> As I understood there is no dedicated maintainer of the optim
> package...

Sadly, that seems to be the state of many packages... :-(

> Eugeniy E. Mikhailo commented on this new patch saying he doesn't see
> anything wrong with this one. So can I commit it?

By all means. I think it looks like a nice clean-up.

Søren

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T03:24:23Z

--- Mer 16/12/09, Jaroslav Hajek <highegg@...> ha scritto:

>
> You see, most of the convenience of working with a mailing
> list comes
> from the capabilities of existing mail clients. Of course,
> if you read
> mail in plain text from command line, any kind of web forum
> must look
> good, but for instance GMail is quite smart and
> customizable, and I
> don't think it's that easy to outmatch.

The other convenience of mailing list is that I can have the
same approach for all the different subscription I need.
As maintainer of octave for cygwin and of most of the needed
math libraries, I need to follow lapack, atlas, graphicsmagick,
glpk, fftw3 and so on.
Lapack and Atlas use a forum approuch and in my humble opinion
are a real waste of time in comparison to the little traffic they handle.

> --
> RNDr. Jaroslav Hajek
>

Marco

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: optim/leasqr.m style changes, diff

2009-12-16T03:15:30Z

On Tue, Dec 15, 2009 at 09:39:13PM +0100, Søren Hauberg wrote:
> man, 14 12 2009 kl. 15:17 +0100, skrev Olaf Till:
> > Anyway I'll keep Matlab compatibility now, but still would like to
> > make the following style changes beforehand:
>
> I think these changes sound just fine (I'm not the maintainer of the
> package, so my opinion isn't all that important).

As I understood there is no dedicated maintainer of the optim
package...

Eugeniy E. Mikhailo commented on this new patch saying he doesn't see
anything wrong with this one. So can I commit it? Afterwards I would
start a new thread with the first "real" change to leasqr.m.

Regards, Olaf

> Personally, I would be
> fine with dropping Matlab compatibility (as your original patch did), I
> just brought up the issue to get an idea of whether or not people were
> dependent on this compatibility.

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T03:04:10Z

On Wed, Dec 16, 2009 at 11:34 AM, Rene Teerwill <teer@...> wrote:

> 2009/12/16 Søren Hauberg <soren@...>:
>> I am a scientist. I don't accept explanations such: "everybody else is
>> doing it, so it must be good". I need a proper rationale to be
>> convinced.
>
> As I said: IMHO Webforums can do everything the mailing list can do
> (which is not a lot) plus a bit more.
> (although I don't know if you can send responses via email)
>
> For me that sounds like a good argument.
>
> A good example how a forum should look like is
> http://www.delphipraxis.net/forum6_vcl+winforms+controls.html
> (a german developer forum, quite active, this subtree has 180.000
> entries, the complete tree has 990.000 entries and is still quick,
> easy to use and good searchable)
>

Nice. But besides the search capabilities, I see no big advantages.
Personal data & website, private messages, that's all easily doable in
a mailing list as well.

I see no "tree view" option. Can you reply to a particular message?
Newcomers are required to register, just like with a mailing list.

Can you filter messages automatically, give them labels, mark them as
important, etc?

You see, most of the convenience of working with a mailing list comes
from the capabilities of existing mail clients. Of course, if you read
mail in plain text from command line, any kind of web forum must look
good, but for instance GMail is quite smart and customizable, and I
don't think it's that easy to outmatch.

--
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev

Re: Replace mailing list with webforum?

2009-12-16T02:34:02Z

2009/12/16 Søren Hauberg <soren@...>:
> I am a scientist. I don't accept explanations such: "everybody else is
> doing it, so it must be good". I need a proper rationale to be
> convinced.

As I said: IMHO Webforums can do everything the mailing list can do
(which is not a lot) plus a bit more.
(although I don't know if you can send responses via email)

For me that sounds like a good argument.

A good example how a forum should look like is
http://www.delphipraxis.net/forum6_vcl+winforms+controls.html
(a german developer forum, quite active, this subtree has 180.000
entries, the complete tree has 990.000 entries and is still quick,
easy to use and good searchable)

>I am definitely willing to consider the switch, but as
>I've said before: the reasons for making a switch, should be good.

Ok, I do understand.
Let's put it the other way round: What functionality do you need?

Btw: I'm mainly talking about the -help - group.
I have no insight in the -core-groups, but I suppose for
developer-communication you will use Mantis or some other bugtracker
anyway.

------------------------------------------------------------------------------
This SF.Net email is sponsored by the Verizon Developer Community
Take advantage of Verizon's best-in-class app development support
A streamlined, 14 day to market process makes app distribution fast and easy
Join now and get one step closer to millions of Verizon customers
http://p.sf.net/sfu/verizon-dev2dev
_______________________________________________
Octave-dev mailing list
Octave-dev@...
https://lists.sourceforge.net/lists/listinfo/octave-dev

From forum: octave-dev