RMS error and Mean absolute error from text files

Question

Muhammad Usman Saleem 2016-3-28

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/275765-rms-error-and-mean-absolute-error-from-text-files

编辑： dpb 2016-4-1

There are 24 text files name from interpot_linear_00.txt to interpot_linear_24.txt in a folder. Each file consist on three columns( First is latitude, second is longitude and third column is temperature). I want to check best of my interpolation technique for 324 samples of temperature.

That why i randomly select 30% of samples and gives them -9999.0. Now using scatter data interpolation i consider -9999.0 as bad data values and based upon my 70% remaining samples interpolate -9999.0. After interpolating values of -9999.0 i cross check its value with its original one and calculate overall R.M.S error and Absolute mean error for each text file.

Here i am just try, but little task to remains which need assistance as in my comments

methods = {'natural'};
S = dir('interpot_linear_*.txt');
N = sort({S.name});
for K = 1 : length(N)
  infile = N{K};
  whichfile = sscanf(infile, 'interpot_linear_%c%c');
      % Load the data
      data = load(infile);
    % separate the data columns, just to make the code clear
      Lat = data(:,1); % Column 1 is Latitude
      Lon = data(:,2); % Column 2 is Longitude
      Tmp = data(:,3); % Column 3 is Temperature
     % Creating 30% of sample as -9999.0
      nr=round(n*0.3);
      TE=Tmp(randperm(n,nr));
      good_temp = find(TE);
      TE(:)=-9999.000;
      %makes another temperature column (say Tmp1) with same indices of randomly
      %seclected but -9999.0 value.
      % Find the "good" data points. It should be 70% of remaing samples?
      good_temp = find(Tmp > -9999.000);
      % find the "bad" data points
      bad_temp = find(Tmp == -9999.000);
      for midx = 1 :length(methods) 
        method = methods{midx};
        outfile = ['interpot_' method '_' whichfile '.txt'];
        % creating vector 
        T = scatteredInterpolant(Lat(good_temp), Lon(good_temp), Tmp(good_temp), method);
        % use the interpolation object to interpolate temperature values
        interp_values = T(Lat(bad_temp), Lon(bad_temp));
        % replace the bad values with the interpolated values
        Tmp(bad_temp) = interp_values;
        % Here i will calculate RMS error with formula for each file
        RMS=sqrt(mean(Tmp(:).^2 - Tmp_new(:).^2));
        %Here i am calculating Mean Absolute Error for each file with formula
        MAE=sum(abs(Tmp(:)-Tmp_new(:)))/n;
        % Here i will export results in a excel file.
      end
  end      %files

After calculating RMS error and AM error for 24 text file. I want summery statistics in excel file( File no, RMS, AME etc.) My some text files has been attached with this post.

Please help!

1 个评论
显示 -1更早的评论隐藏 -1更早的评论

Stephen23 2016-3-28

编辑：Stephen23 2016-3-28

The code in the question is derived from these answers (questions also from same OP)

http://www.mathworks.com/matlabcentral/answers/265845-1d-interpolation-from-multiple-text-files

http://www.mathworks.com/matlabcentral/answers/270450-scatter-data-interpolation-from-text-files-in-matlab

http://www.mathworks.com/matlabcentral/answers/272129-bilinear-interpolation-of-2d-matrix

请先登录，再进行评论。

请先登录，再回答此问题。

Answer 1

dpb 2016-3-28

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/275765-rms-error-and-mean-absolute-error-from-text-files#answer_215326

编辑：dpb 2016-3-28

在 MATLAB Online 中打开

...
% Creating 30% of sample as -9999.0
nr=round(n*0.3);
TE=Tmp(randperm(n,nr));
good_temp = find(TE);
...

In the code above you've not defined n which, it appears, should be length of the vectors or

n=size(data,1);

Then randperm will produce a pseudo-random sample of those indices as you've used it. BUT, what you've called "good_temp" is that 30% subset, not the remaining 70% so you're backwards on which group is which it would appear.

BTW, using just the indices is simpler; you don't need find there...

idx30=randperm(n,nr);  % index of 30% for substitution
tmp(idx30)=nan;        % use Nan instead of "magic number" for missing value

The remaining 70% are those which aren't members of the above set. It's simpler to use logical addressing than absolute for such cases; to generate use

ibad=ismember(1:n,idx30);  % logical vector of 30% bad locations
igood=~ibad;               % complementary "good" locations vector

Using the above index arrays, you don't need to do searching for the missing values; simply use the indexing array for the subset desired in expressions or, the isnan function or the builtin functions such as nanmean and friends which automagically handle NaN as missing value (latter may require Statistics Toolbox depending on release).

9 个评论
显示 7更早的评论隐藏 7更早的评论

Muhammad Usman Saleem 2016-3-28

在 MATLAB Online 中打开

Thanks @dpb...

Here i am trying this to one text file with name interpot_linear_00.txt

% Load the data
data = load('interpot_linear_00.txt');
  % separate the data columns, just to make the code clear
    Lat = data(:,1); % Column 1 is Latitude
    Lon = data(:,2); % Column 2 is Longitude
    Tmp = data(:,3); % Column 3 is Temperature
     % Creating 30% of sample as -9999.0
       n=size(data,1);
      nr=round(n*0.3);
      idx30=randperm(n,nr);
      Tmp(idx30)=-9999.0;
      % Find the "good" data points
      good_temp = find(Tmp > -9999.000);
       % find the "bad" data points
      bad_temp = find(Tmp == -9999.000);
       % creating vector 
        T = scatteredInterpolant(Lat(good_temp), Lon(good_temp), Tmp(good_temp), 'natural');
      % use the interpolation object to interpolate temperature values
        interp_values = T(Lat(bad_temp), Lon(bad_temp));
        % replace the bad values with the interpolated values
        Tmp(bad_temp) = interp_values;
        %recall oringinal temperature column
      Tmp_original=data(:,3);
      % Here calculate RMS error
      RMS=sqrt(mean(Tmp_original(:).^2 - Tmp(:).^2));
      % Here calculate mean absolute error
      MAE=sum(abs(Tmp_original(:)-Tmp(:)))/n;
      % Here calculate correlation coefficient
      correl=corrcoef(Tmp_original(:),Tmp(:));
      %making results matrix
      results=[RMS;MAE;correl];
      %here i am exporting my results
      xlswrite(Statistics,results);

Please check it and tell me whether this approach is fine??

As there are 24 text files. I want to export final excel file with summery of statistics of each file. Like first column is file no, second column is RMS error, third column is MAE and fourth column of correl.

One excel file will contains all these information??

Muhammad Usman Saleem 2016-3-30

编辑：Muhammad Usman Saleem 2016-3-30

在 MATLAB Online 中打开

In understanding of "you'll need to either save all the summary results in the loop over the files and then write those to an external file (Excel or other makes on difference) after the loop completes or write each result to the same file during the loop."

Here i am doing

   S = dir('Output_*.txt');
  N = sort({S.name});
for K = 1 : length(N)
  infile = N{K};
  whichfile = sscanf(infile, 'Output_%c%c');
% Load the data
data = load(infile);
  % separate the data columns, just to make the code clear
    Lat = data(:,1); % Column 1 is Latitude
    Lon = data(:,2); % Column 2 is Longitude
    Tmp = data(:,3); % Column 3 is Temperature
    %Creating 30% of sample as -9999.0
     n=size(data,1);
    nr=round(n*0.3);
    idx30=randperm(n,nr);
    Tmp(idx30)=-9999.0;
     % Find the "good" data points
    good_temp = find(Tmp > -9999.000);
     % find the "bad" data points
    bad_temp = find(Tmp == -9999.000);
      % creating vector 
      T = scatteredInterpolant(Lat(good_temp), Lon(good_temp), Tmp(good_temp), 'natural');
    % use the interpolation object to interpolate temperature values
      interp_values = T(Lat(bad_temp), Lon(bad_temp));
      % replace the bad values with the interpolated values
      Tmp(bad_temp) = interp_values;
      %recall oringinal temperature column
    Tmp_original=data(:,3);
    % Here calculate RMS error
    RMS=sqrt(mean(Tmp_original(:).^2 - Tmp(:).^2));
    % Here calculate mean absolute error
    MAE=sum(abs(Tmp_original(:)-Tmp(:)))/n;
    % Here calculate correlation coefficient
    correl=corrcoef(Tmp_original(:),Tmp(:));
    results=[RMS';MAE';correl']';
    for ii=1:size(results,1)
        xlswrite(Statistics,results);
    end
end

Please correct it, in reward i shall accept your kind answer?

Muhammad Usman Saleem 2016-3-31

编辑：Muhammad Usman Saleem 2016-3-31

在 MATLAB Online 中打开

@dpb for breaking such silence...

Yes you are absolutely right. There was a wrong file in that folder which was empty. I have delete it. Now when i try make a matrix of each results through

results=[RMS';MAE';correl']';

and even

rep=[RMS(:);MAE(:);correl(:)];

It is storing statistics of only 24 no text file(last one) skipping results of others . How can i store results of each text file in results and at end 24 * 3 matrix will be summery statistics of all files and then how can i export to results in excel? Please

dpb 2016-3-31

编辑：dpb 2016-4-1

No, it isn't "skipping" anything; your assignment is simply that--an assignment of the current values to a variable. It's doing precisely what you told it to.

Read the documentation on <elementary-matrices-and-arrays> particular the section and examples <Creating-and-concatenating-matrices>

ADDENDUM

HINT: As I described for you quite some time ago, you must either

preallocate room for the output array and then populate that array with an index expression to store the results in subsequent rows, or
initialize the array variable to the empty vector by '[]' and then dynamically concatenate each element into that array within the loop.

As is, as I've pointed out also multiple times, you're simply assigning the results to the same array variable in its entirety over and over, each pass thru the loop consequently overwriting the previous loop's result. Consequently, the symptom you see of the last pass only being the one result left standing.

Also, once you have finally worked this out, you'll want to move the instruction to write that array to an external file to after the loop; else't you'll be writing the sequential pieces over and over, too...

OR (note well) instead of creating an array, you could simply write each result as it is computed. If this is a sequential file, nothing more need than to ensure include the '\n' in the format string. If indeed must use (ugh!) Excel, then you'll have to (as again noted long ago) create the addressing expression for each write to avoid doing the same thing there you're now doing in memory. For that reason, this option is less palatable than the previous but include it for completeness.

请先登录，再进行评论。

Answer 2

Walter Roberson 2016-4-1

1
链接

此回答的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/275765-rms-error-and-mean-absolute-error-from-text-files#answer_215877

I have glanced at your question and I see that Duane is handling it perfectly well. You do not need me for it.

3 个评论
显示 1更早的评论隐藏 1更早的评论

Walter Roberson 2016-4-1

I did, several hours before.

Muhammad Usman Saleem 2016-4-1

let the story to be hear soon. If you help me to find correction in the other question. I hope and sure for this question you have did, several hours before.

请先登录，再进行评论。