Input txt. file of complex numbers

I have large data text files from COMSOL that contain multiple rows and columns of complex numbers. I have tried importing them into matlab as the files are quite large, but all the data is presented in a single column. eg:

-12.475000000000005       -1.975                   2.5658824311275345E-6+4.421192310257837E-6i       -4.688910807327458E-7-9.893058421633841E-8i       5.549150260222212E-6-2.1496631444278723E-5i       -2.7187186095082164E-9+1.1189876165621303E-8i   

I have tried to include Space as a column delimiter but it gives me the message: "com.mathworks.jmi.MatlabException: Arrays have incompatible sizes for this operation". How can I import this data properly?

[dpb]

Attach a small section of the file itself -- we can't tell what's in it otherwise...but, is it by chance a fixed-width file, not comma or tab delimited?

[Sam Hurrell]

I have inputed files from COMSOL to Matlab in the same way before and didn't have this problem. The only difference with this file is that it contained complex number

[dpb]

MAJOR weakness in the C formatted i/o library and incomprehensible to me that Mathworks hasn't addressed it in a scientific package -- there is no provision whatever for complex variables in the fscanf formatting -- and TMW hasn't done anything to help with all the new readXXX family, either.

And, the folks who wrote the file didn't do you any favors by including the real x,y coordinates on the same record as the first of the variables so will have to parse them out individually as well.

At least there is the header

% Dimension:          2
% Nodes:              20000
% Expressions:        60

that you can read to determine there are precisely two dimensions to read and then apparently "Expressions" is the number of values;

I didn't look at the whole file, but I then presume there are "Nodes" numbers of those repeated.

With some trial and tribulations, textscan or fscanf will be able to deal with this; or it might be more conducive albeit undoubtedly slower for a regular expressions conversion.

Might do a search and see if somebody has a FEX submission or there's a posted MATLAB function written already to parse a COMSOL file first, though...

Answer 1

[Johannes Hougaard]

As far as I can tell you can use the readmatrix function to read the file if you use the 'NumHeaderLines' option for the import.

COMSOL = readmatrix("COMSOL.TXT",'NumHeaderLines',9);

[dpb]

Well, superficially, maybe, but in reality it got really, really confused --

>> d=readmatrix('COMSOL.txt','NumHeaderLines',9);
> d(:,1:6)
ans =
 -12.475000000000005                 NaN                 NaN                 NaN                 NaN                 NaN
 -12.425000000000004                 NaN                 NaN                 NaN                 NaN                 NaN
 -12.375000000000004                 NaN                 NaN                 NaN                 NaN                 NaN
 -12.325000000000003                 NaN                 NaN                 NaN                 NaN                 NaN
 -12.275000000000002                 NaN                 NaN                 NaN                 NaN                 NaN
 -12.225000000000005                 NaN                 NaN                 NaN                 NaN                 NaN
>> 

I thought at first it was ok if did

>> d=d(isfinite(d));
>> d=reshape(d,6,[]);
>> d(1:2,1:4)
ans =
-12.475000000000005 + 0.000000000000000i -1.975000000000000 + 0.000000000000000i  0.000002565882431 + 0.000004421192310i -0.000000486622006 - 0.000000103889320i
-12.425000000000004 + 0.000000000000000i -1.975000000000000 + 0.000000000000000i  0.000002634471990 + 0.000004533952889i -0.000000506108402 - 0.000000108803666i
>> 

but if compare to the input file, the first two match up with the nodes although the two reals are stored as complex, and the first element of the complex variables is correct, but the second in the first record actually matches up to the second element in the second record -- the file pointer has gotten all confused about where are in the conversion process; probably owing to there being two reals first.

Unfortunately, while it did something and did return a complex variable, one can't tell it about there being the two reals first -- that's a real pain in reading the form in which the file was created -- would be much easier with that record by itself.

And, even if it were, there's no 'complex' data type/class supported in the import options object.

[dpb]

Thought -- this might be a place for https://www.mathworks.com/help/matlab/ref/memmapfile.html but I don't have time to play any longer right now...

[dpb]

readmatrix isn't up to the task even if you remove the two reals from the beginning of each record -- not that that's even practical for a real solution but was just curious --

>> d=readmatrix('COMSOL1.txt','NumHeaderLines',9);
>> whos d
  Name      Size             Bytes  Class     Attributes

  d         6x350            33600  double    complex   

>> d(:,1:4)
ans =
   1.0e-05 *
  0.256588243112753 + 0.442119231025784i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
  0.263447199023142 + 0.453395288888662i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
  0.270764788515978 + 0.464910655188620i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
  0.278352632788851 + 0.476980423334000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
  0.286256553257367 + 0.489424192143307i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
  0.294435828095543 + 0.502476218324594i                NaN + 0.000000000000000i                NaN + 0.000000000000000i                NaN + 0.000000000000000i
>>

It still got confused after the first record for some reason; that it is space-delimited but not exactly fixed-width may be part of the problem althoug in general C will just "eat" whitespace...

>> fid=fopen('COMSOL1.txt','r');
>> for i=1:9,fgetl(fid);end
>> l=fgetl(fid);
>> find(l==9)
ans =
  1×0 empty double row vector
>> find(l==32,5)
ans =
    44    45    46    47    48
>> fid=fclose(fid);
>> 

But it isn't tab-delimited; just lots of blanks...

[Johannes Hougaard]

What MATLAB version are you on? When I run the code I get:

>>comsol = readmatrix("COMSOL.txt",'NumHeaderLines',9);
>> whos('comsol')
  Name        Size            Bytes  Class     Attributes

  comsol      6x62             5952  double    complex  
  >> comsol(:,3:6)

ans =

   1.0e-04 *

   0.0257 + 0.0442i  -0.0047 - 0.0010i   0.0555 - 0.2150i  -0.0000 + 0.0001i
   0.0263 + 0.0453i  -0.0049 - 0.0010i   0.0567 - 0.2206i  -0.0000 + 0.0001i
   0.0271 + 0.0465i  -0.0051 - 0.0011i   0.0580 - 0.2265i  -0.0000 + 0.0001i
   0.0278 + 0.0477i  -0.0053 - 0.0011i   0.0593 - 0.2326i  -0.0000 + 0.0001i
   0.0286 + 0.0489i  -0.0055 - 0.0012i   0.0607 - 0.2390i  -0.0000 + 0.0001i
   0.0294 + 0.0502i  -0.0057 - 0.0013i   0.0622 - 0.2456i  -0.0000 + 0.0001i

[Johannes Hougaard]

...but if you'd prefer getting rid of readmatrix and using file level read options that's doable too

>> fid = fopen("COMSOL.txt",'r');
comsol = cell(0,0);
while ~feof(fid)
    thisline = fgetl(fid);
    comsol = vertcat(comsol,{thisline}); %#ok<AGROW> %accept that it's a growing variable although slightly slow
end
fclose(fid);
incl = false(size(comsol));
for ii = 1:length(comsol)
    if ~strncmp(comsol{ii},'%',1)
        comsol{ii} = str2num(comsol{ii}); %#ok<ST2NM> % str2num does the trick, str2double doesn't work on a formatted string
        incl(ii) = true;
    end
end
numericalcomsol = cell2mat(comsol(incl,:));
clear ans ii fid thisline incl
whos
  Name                  Size            Bytes  Class     Attributes

  comsol               15x1             14272  cell                
  numericalcomsol       6x62             5952  double    complex   
>> numericalcomsol(:,3:6)

ans =

   1.0e-04 *

   0.0257 + 0.0442i  -0.0047 - 0.0010i   0.0555 - 0.2150i  -0.0000 + 0.0001i
   0.0263 + 0.0453i  -0.0049 - 0.0010i   0.0567 - 0.2206i  -0.0000 + 0.0001i
   0.0271 + 0.0465i  -0.0051 - 0.0011i   0.0580 - 0.2265i  -0.0000 + 0.0001i
   0.0278 + 0.0477i  -0.0053 - 0.0011i   0.0593 - 0.2326i  -0.0000 + 0.0001i
   0.0286 + 0.0489i  -0.0055 - 0.0012i   0.0607 - 0.2390i  -0.0000 + 0.0001i
   0.0294 + 0.0502i  -0.0057 - 0.0013i   0.0622 - 0.2456i  -0.0000 + 0.0001i