Thread: Help needed to perform bootstrap with matlab

I have been given an m-file that amongst other things does a bootstrap calculation but I seem to be hitting a brick wall.

Here is the file I am using:

[nz,ntv] = size(obs);
% means
disp(' ')
mobs = mean(obs);
disp (['mean observed values = ' num2str(mobs)])
mest = mean(est);
disp (['mean estimated values = ' num2str(mest)])
% standart deviations
sdobs = std(obs);
disp (['std dev observed values = ' num2str(sdobs)])
sdest = std(est);
disp (['std dev estimated values = ' num2str(sdest)])
% assessment of goodness-of-fit
% correlation and coefficient of determination (explained variance)
[corre,p] = corrcoef([obs est]);R2=corre.^2;
disp (['R2 corre^2 = ' num2str(R2(1,2)) ' p = ' num2str(p(1,2))])
% mean absolute error
imae = sum(abs(obs-est))/ntv;
disp (['mae = ' num2str(imae)])
% root mean square error
rmse = sqrt(sum((obs-est).^2)/ntv);
disp (['rmse = ' num2str(rmse)])
disp(' ')
% Nash-Sutcliffe coeff of efficiency
E = 1-(sum((obs-est).^2)/sum((obs-mobs).^2));
disp (['N-Sut efficiency (E) = ' num2str(E)])
stat = bootstrp(500,@Nash_Sut,obs,est);
[a b]=ci95(stat);
disp ([ 'E 95% ci = ' num2str(a) ' ' num2str(b)])
disp (' ');clear stat
% log Nash-Sutcliffe coeff of efficiency
Elog = 1-(sum((log(obs)-log(est)).^2)/sum((log(obs)-log(mobs)).^2));
disp (['Elog = ' num2str(Elog)])
stat = bootstrp(500,@log_Nash_Sut,obs,est);
[a b]=ci95(stat);
disp ([ 'E 95% ci = ' num2str(a) ' ' num2str(b)])
disp (' ');clear stat
% Modified Nash-Sutcliffe coeff of efficiency
E1 = 1-(sum(abs(obs-est))/sum(abs(obs-mobs)));
disp (['N-Sut efficiency modified (E1) = ' num2str(E1)])
stat = bootstrp(500,@m_Nash_Sut,obs,est);
[a b]=ci95(stat);
disp ([ 'E1 95% ci = ' num2str(a) ' ' num2str(b)])
disp (' '); clear stat
% index of agreement
d = 1-(sum((obs-est).^2)/sum((abs(obs-mobs)+abs(obs-mobs)).^2));
disp (['Index of agreement (d) = ' num2str(d)])
stat = bootstrp(500,@I_agree,obs,est);
[a b]=ci95(stat);
disp ([ 'd 95% ci = ' num2str(a) ' ' num2str(b)])
disp (' '); clear stat
% Modified index of agreement
d1 = 1-(sum(abs(obs-est))/sum((abs(obs-mobs)+abs(obs-mobs))));
disp (['Index of agreement modified (d1) = ' num2str(d1)])
stat = bootstrp(500,@m_I_agree,obs,est);
[a b]=ci95(stat);
disp ([ 'd1 95% ci = ' num2str(a) ' ' num2str(b)])
disp (' '); clear stat

and this is the result:

mean observed values = 17.9477
mean estimated values = 17.7683
std dev observed values = 6.4146
std dev estimated values = 2.1859
R2 corre^2 = 0.034055 p = 0.32894
mae = 152.28
rmse = 34.3907

N-Sut efficiency (E) = 0.0088419
??? Error using ==> feval
Undefined command/function 'Nash_Sut'.

Error in ==> bootstrp at 77
thetafit = feval(bootfun,varargin{:});

Error in ==> goodfit2 at 53
stat = bootstrp(500,@Nash_Sut,obs,est);

Can anyone help me resolve the problem? It seems that matlab does not like Nash_sut and the rest as a Bootfun.

Secondly, and more to save time, how do I modify the m-file so that instead of running this for every obs and est file I have (there are 50+ of each), I could put them all into one file each ie in columns and get the outputs for each individual column?

Originally Posted by Kean

I have been given an m-file that amongst other things does a bootstrap calculation but I seem to be hitting a brick wall.

Here is the file I am using:


and this is the result:


Can anyone help me resolve the problem? It seems that matlab does not like Nash_sut and the rest as a Bootfun.

Secondly, and more to save time, how do I modify the m-file so that instead of running this for every obs and est file I have (there are 50+ of each), I could put them all into one file each ie in columns and get the outputs for each individual column?

offending line is:

stat = bootstrp(500,@Nash_Sut,obs,est);

Where in Nash_Sut defined, and what is it?? If it is a function handel for an anonymous function then it does not need the @.

CB

Hi CB,

from what I know nash_sut is the function above ie: E = 1-(sum((obs-est).^2)/sum((obs-mobs).^2)).

The idea is that by bootstraping one can determine the significance of E. How would I be able to correct the problem/rewrite the code to fix this?

Kean

Originally Posted by Kean

Hi CB,

from what I know nash_sut is the function above ie: E = 1-(sum((obs-est).^2)/sum((obs-mobs).^2)).

The idea is that by bootstraping one can determine the significance of E. How would I be able to correct the problem/rewrite the code to fix this?

Kean

Then what is the @ doing in that statement?

CB

You got me! I have no idea, i am a relative newbie to matlab and definitly never used bootstrp before.

So if I want the bootfun to be the function mentioned how do I enter it into the bootstrp statement.

I tried to enter the entire function but that did not work, should I create a new m-file with the function, and then enter that m-file?