leasqr and anonymous functions

[Nose Nada]

Hello there!

I have defined an anonymous function Error with three parameters (SD(1:3)) and an independent variable p:

octave:1> SDc = 0.0001; R=1000;
octave:2> Error = @(SD,p)(SDc^2 + ((1/(R-1) + p).^2)*(SD(1)^2) + (((p./((2*(SD(3)^1.5) + 0.6))).*log(1+(10^SD(3)).*(1-p))).^2).*(SD(2)^2));

It worked fine on evaluation:

octave:3> Error([0.01 0.01 1], 0:0.1:1)
ans =

   1.0100e-08   1.8144e-06   6.9068e-06   1.4827e-05   2.5052e-05   3.6983e-05   4.9925e-05   6.3081e-05   7.5597e-05   8.6947e-05   1.0021e-04

However, when I tried to fit it to experimental data using 'leasqr', it fails:

octave:4> leasqr((0:0.1:1)',(0.001:0.001:0.011)', [0.01 0.01 1], Error)
error: operator -: nonconformant arguments (op1 is 11x1, op2 is 3x1)
error: evaluating binary operator `-' near line 182, column 9
error: evaluating binary operator `.*' near line 182, column 5
error: evaluating assignment expression near line 182, column 2
error: called from `leasqr' in file `/usr/share/octave/packages/optim-1.0.0/leasqr.m'


What could be the problem?
Thanks in advance,
Omar

[Olaf Till]

On Tue, Apr 29, 2008 at 07:58:48PM -0700, Nose Nada wrote:

>
> Hello there!
>
> I have defined an anonymous function Error with three parameters (SD(1:3))
> and an independent variable p:
>
> octave:1> SDc = 0.0001; R=1000;
> octave:2> Error = @(SD,p)(SDc^2 + ((1/(R-1) + p).^2)*(SD(1)^2) +
> (((p./((2*(SD(3)^1.5) + 0.6))).*log(1+(10^SD(3)).*(1-p))).^2).*(SD(2)^2));
>
> It worked fine on evaluation:
>
> octave:3> Error([0.01 0.01 1], 0:0.1:1)
> ans =
>
>    1.0100e-08   1.8144e-06   6.9068e-06   1.4827e-05   2.5052e-05  
> 3.6983e-05   4.9925e-05   6.3081e-05   7.5597e-05   8.6947e-05   1.0021e-04
>
> However, when I tried to fit it to experimental data using 'leasqr', it
> fails:
>
> octave:4> leasqr((0:0.1:1)',(0.001:0.001:0.011)', [0.01 0.01 1], Error)
> error: operator -: nonconformant arguments (op1 is 11x1, op2 is 3x1)
> error: evaluating binary operator `-' near line 182, column 9
> error: evaluating binary operator `.*' near line 182, column 5
> error: evaluating assignment expression near line 182, column 2
> error: called from `leasqr' in file
> `/usr/share/octave/packages/optim-1.0.0/leasqr.m'
>
>

Have not looked thoroughly through your function Error, but it seems
to have its arguments in the wrong order. Although the second argument
is named 'p', the function seems to treat it as the data, not the
parameters. leasqr calles Error with p = [0.01 0.01 1], so error
returns only 3 values instead of 11.

Olaf
_______________________________________________
Help-octave mailing list
Help-octave@...
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Nose Nada]

Thanks Olaf!
I changed the definition of the anonymous function from @(SD,p) to @(p,SD) and then leasqr run smoothly. However, all the output was in imaginary numbers and they must be real. I tried to find if among the input arguments of leasqr there was the possibility to define lower and upper bounds for the parameters values, as it is the case in Matlab, but I found nothing. Do you know if in Octave there is some minimizer of sum of squared residuals that allows the definition of parameter ranges? Must I use some general minimizer (e.g. fminbnd  or samin) with that possibility and then define the sum of residual squares by myself?
I look forward for some help.
Thanks in advance,
Omar





On Tue, Apr 29, 2008 at 07:58:48PM -0700, Nose Nada wrote:
>
> Hello there!
>
> I have defined an anonymous function Error with three parameters (SD(1:3))
> and an independent variable p:
>
> octave:1> SDc = 0.0001; R=1000;
> octave:2> Error = @(SD,p)(SDc^2 + ((1/(R-1) + p).^2)*(SD(1)^2) +
> (((p./((2*(SD(3)^1.5) + 0.6))).*log(1+(10^SD(3)).*(1-p))).^2).*(SD(2)^2));
>
> It worked fine on evaluation:
>
> octave:3> Error([0.01 0.01 1], 0:0.1:1)
> ans =
>
>    1.0100e-08   1.8144e-06   6.9068e-06   1.4827e-05   2.5052e-05  
> 3.6983e-05   4.9925e-05   6.3081e-05   7.5597e-05   8.6947e-05   1.0021e-04
>
> However, when I tried to fit it to experimental data using 'leasqr', it
> fails:
>
> octave:4> leasqr((0:0.1:1)',(0.001:0.001:0.011)', [0.01 0.01 1], Error)
> error: operator -: nonconformant arguments (op1 is 11x1, op2 is 3x1)
> error: evaluating binary operator `-' near line 182, column 9
> error: evaluating binary operator `.*' near line 182, column 5
> error: evaluating assignment expression near line 182, column 2
> error: called from `leasqr' in file
> `/usr/share/octave/packages/optim-1.0.0/leasqr.m'
>
>
Have not looked thoroughly through your function Error, but it seems
to have its arguments in the wrong order. Although the second argument
is named 'p', the function seems to treat it as the data, not the
parameters. leasqr calles Error with p = [0.01 0.01 1], so error
returns only 3 values instead of 11.

Olaf
_______________________________________________
Help-octave mailing list
[e-mail]
https://www.cae.wisc.edu/mailman/listinfo/help-octave

[Nose Nada]

I have solve the optimization problem defining the sum of squared residuals (between the data and the model 'Error') as an independent function script and then minimizing that function using samin which allows for constraining the parameter range. It seems to be running smoothly.
Thanks for the help, specially to Olaf.
Omar
 

Nose Nada wrote:

Thanks Olaf!
I changed the definition of the anonymous function from @(SD,p) to @(p,SD) and then leasqr run smoothly. However, all the output was in imaginary numbers and they must be real. I tried to find if among the input arguments of leasqr there was the possibility to define lower and upper bounds for the parameters values, as it is the case in Matlab, but I found nothing. Do you know if in Octave there is some minimizer of sum of squared residuals that allows the definition of parameter ranges? Must I use some general minimizer (e.g. fminbnd  or samin) with that possibility and then define the sum of residual squares by myself?
I look forward for some help.
Thanks in advance,
Omar





On Tue, Apr 29, 2008 at 07:58:48PM -0700, Nose Nada wrote:
>
> Hello there!
>
> I have defined an anonymous function Error with three parameters (SD(1:3))
> and an independent variable p:
>
> octave:1> SDc = 0.0001; R=1000;
> octave:2> Error = @(SD,p)(SDc^2 + ((1/(R-1) + p).^2)*(SD(1)^2) +
> (((p./((2*(SD(3)^1.5) + 0.6))).*log(1+(10^SD(3)).*(1-p))).^2).*(SD(2)^2));
>
> It worked fine on evaluation:
>
> octave:3> Error([0.01 0.01 1], 0:0.1:1)
> ans =
>
>    1.0100e-08   1.8144e-06   6.9068e-06   1.4827e-05   2.5052e-05  
> 3.6983e-05   4.9925e-05   6.3081e-05   7.5597e-05   8.6947e-05   1.0021e-04
>
> However, when I tried to fit it to experimental data using 'leasqr', it
> fails:
>
> octave:4> leasqr((0:0.1:1)',(0.001:0.001:0.011)', [0.01 0.01 1], Error)
> error: operator -: nonconformant arguments (op1 is 11x1, op2 is 3x1)
> error: evaluating binary operator `-' near line 182, column 9
> error: evaluating binary operator `.*' near line 182, column 5
> error: evaluating assignment expression near line 182, column 2
> error: called from `leasqr' in file
> `/usr/share/octave/packages/optim-1.0.0/leasqr.m'
>
>
Have not looked thoroughly through your function Error, but it seems
to have its arguments in the wrong order. Although the second argument
is named 'p', the function seems to treat it as the data, not the
parameters. leasqr calles Error with p = [0.01 0.01 1], so error
returns only 3 values instead of 11.

Olaf
_______________________________________________
Help-octave mailing list
[e-mail]
https://www.cae.wisc.edu/mailman/listinfo/help-octave