What is the fastest way to extract data from a huge text file?

I have a text file like this:

     1.0            IONOSPHERE MAPS     GNSS                IONEX VERSION / TYPE
ADDNEQ2 V5.3        AIUB                03-JUL-14 20:57     PGM / RUN BY / DATE 
CODE'S GLOBAL IONOSPHERE MAPS FOR DAY 180, 2014             COMMENT             
Global ionosphere maps (GIM) are generated on a daily basis DESCRIPTION         
(I don't want this part)
.
.
.
(skip 600 lines)
    1                                                      START OF TEC MAP    
  2014     6    29     0     0     0                        EPOCH OF CURRENT MAP
    87.5-180.0 180.0   5.0 450.0                            LAT/LON1/LON2/DLON/H
  154  154  155  155  155  156  156  156  156  155  155  155  154  154  153  153
  152  151  150  149  148  147  146  145  145  144  143  142  141  140  139  139
  138  138  137  137  137  137  136  136  137  137  137  137  137  138  138  139
  139  139  140  140  141  142  142  143  143  144  145  145  146  147  147  148
  149  149  150  151  152  152  153  153  154
    85.0-180.0 180.0   5.0 450.0                            LAT/LON1/LON2/DLON/H
  160  161  162  163  164  164  165  165  165  164  164  163  163  162  161  159
  158  157  155  153  151  149  147  145  143  141  139  138  136  134  133  132
  131  130  130  129  129  129  130  130  131  131  132  133  134  135  136  136
  137  138  139  139  140  140  141  142  142  143  144  145  146  146  148  149
  150  151  153  154  155  157  158  159  160
.
.
.

I have to search for a specific value by entering specific latitude, longitude and time.

I have a function using fopen and fgetl for searching this. The data have a fixed spacing. So, I use strcmp string comparison and isequal to search for the value I want. . . .

Let say, value = search(lat, lon, time)

lat = 85.0; lon = -175; time (UT) = 0;

I will first compare each line getting from fgetl with the string:

2014 6 29 0 0 0 EPOCH OF CURRENT MAP

If matched, then search for 85.0 from the from following line getting by fgetl

85.0-180.0 180.0 5.0 450.0 LAT/LON1/LON2/DLON/H

If matched, store all related data into a vector:

160 161 162 163 164 164 165 165 165 164 164 163 163 162 161 159 158 157 155 153 151 149 147 145 143 141 139 138 136 134 133 132 131 130 130 129 129 129 130 130 131 131 132 133 134 135 136 136 137 138 139 139 140 140 141 142 142 143 144 145 146 146 148 149 150 151 153 154 155 157 158 159 160 (in vector form)

then get the value by specific vector index (corresponding to longitude, index=2 in this example)

. . .

But I have to call this search function for 250,000 times. It will take over 24 hours!!!

How can I do? I cannot change my computer. Thank!! I need your help!!*

PS: the text file is about 12,000 row * 80 column

[dpb]

12000 lines by the 16 values shown is only 1.5 MB of doubles and since they all appear to be small integer values a single will be exact which would only be half that.

Hence, read the whole file into memory and do all your searching internally.

If spacings are, as you say, fixed, you should be able to compute the index directly rather than search.

Use textread to parse the file on input.

[per isakson]

• Especially if it's urgent, state the problem fully and clearly. Then show what you done. See http://meta.stackexchange.com/questions/66377/what-is-the-xy-problem.
• Upload a sample file. Use the paper-clip-button.
• Use numbers rather than words like "huge".
• "time (UT) = 0;" &nbsp How should that user input be used?
• And if it's "urgent" react on comments

[SO]

Sorry for my unclear question. Actually the spacing are fixed for certain parts of the file only.

I have attached the whole text file now and also my search function.

My main function will call this search function in a nested for loop

for lat = 80:-1:-80

    for long = -180:1:180

    (Call the search function 4 time - 4 different set of input)
    end

end

To test the code, sample_search_function(152,0,lat,long)

Thank for your help!

[per isakson]

• "4 time - 4 different set of input" &nbsp is that a third loop? How do the sets differ?
• "time (UT) = 0;" &nbsp How should time be treated?
• how should the result be organized? What should the output of the search function look like?

[per isakson]

• What about "EPOCH OF CURRENT MAP" ?
• "function [gim_tec] = sample_search_function(DOY, UT, lat, long)" &nbsp I don't want to reverse engineer your function.

[SO]

a = sample_search_function(152,0,lat,long)
b = sample_search_function(152,0,lat+2.5,long)
c = sample_search_function(152,0,lat,long+5)
d = sample_search_function(152,0,lat+2.5,long+5)

I will do interpolation using this four values to calculate a single value. (also do inside the second for loop)

• all done in the second loop.
• About the time, I just search for the line containing the string "EPOCH OF CURRENT MAP" and compare the number of specific location of this line (since the location of time is fixed at this line)
• If time matched, then start search for latitude, and then longitude.
• The output of the search function is a single number (integer, e.g. 160, 261 80, 0...)
• This single number will be store in a matrix after finishing the second for loop.

result(row, col) = output_of_search_function
(second for loop, fill all the column in one row; then start another row when next first for loop start)

[SO]

Is this search function slow down my program or the nested for loop?

How can I speed up?

Thank.

[per isakson]

I've made it run with the filename hardcoded

    >> [gim_tec] = sample_search_function( 20, 85, 0 )
    gim_tec =
       164
    >>

"((mod(long,5) ~=0) (long > 180) (long < -180))" gives 71 values rather than 73

    tec_list      1x73              584  double

why not

    ((mod(long,5) ~=0) || (long >= 180) || (long <= -180))

[per isakson]

Why not read the file once and output an intermediate array of the size < 73x70x11> (lon,lat,ut) and store that in a mat-file. Then you use that array in the your loops.

Reading and parsing the file once will increase speed.

[SO]

• Since the data from the text file only got -180 to 180 longitude with 5 degree interval.
• Input Longitude cannot be completely divided by 5 , or larger than 180 or smaller than -180, will not be accepted. This If condition is correct.

[SO]

There are lots of unwanted data.

I want these kind of data only:

154  154  155  155  155  156  156  156  156  155  155  155  154  154  153  153
152  151  150  149  148  147  146  145  145  144  143  142  141  140  139  139
138  138  137  137  137  137  136  136  137  137  137  137  137  138  138  139
.
.

• Creating a < 73x70x11> (lon,lat,ut) array is a possible solution. But I have to ignore the unwanted data and re-organize the structure of the data (e.g. one row for all data within the same latitude instead of using 5 rows).Any faster way to extract wanted data and re-organize them instead of using fgetl?

Thank for your advice.!!

Answer 1

[per isakson]

Now I'm done:

• less than a tenth of a second to read and parse the sample file (with the file in the system cache)
• less than a tenth of a millisecond to retrieve one value
• the array ION is half a MB. Make ION uint8 to save memory - if needed.
• 62196 values retrieved from the sample file.

You add tests and comments!

    >> tic,ION = cssm();toc
    Elapsed time is 0.074765 seconds.
    >> sum(not(isnan(ION(:))))
    ans =
           62196
    >> whos ION
      Name       Size                Bytes  Class     Attributes

      ION       73x71x12            497568  double            

    >> ION(lon2ix(0),lat2ix(85),ut2ix(20))
    ans =
       164  

    >> tic,ION(lon2ix(0),lat2ix(85),ut2ix(20));toc
    Elapsed time is 0.000067 seconds.

compared to

    >> tic, [gim_tec] = sample_search_function( 20, 85, 0 ), toc
    gim_tec =
       164
    Elapsed time is 0.265756 seconds.

where

    function    ION = cssm()
    str = fileread( 'c:\m\cssm\CODG1520.txt' );
    ca1 = regexp( str, '(?<=START OF TEC MAP).+?(?=END OF TEC MAP)', 'match' );
    ION = nan( 73, 70, 11 );
    lat2ix  = @(lat) round((lat+87.5)/2.5)+1;
    lon2ix  = @(lon) round((lon+180)/5.0)+1; %#ok<NASGU>
    ut2ix   = @(ut)  round(ut/2)+1;
    for jj = 1 : length( ca1 )

        buf = regexp( ca1{jj}, '\n', 'split', 'once' );
        buf = regexp( buf{2} , '\n', 'split', 'once' );
        ut  = textscan( buf{1}, '%*f%*f%*f%f%*[^\n]' );
        ut  = ut{1};
        ca2 = regexp( buf{2}, 'LAT/LON1/LON2/DLON/H', 'split' );
        pos = ca2{1};
        for kk = 2 : length( ca2 )
            lat = textscan( pos,'%f%*[^\n]' ); 
            lat = lat{1};
            num = sscanf( ca2{kk}(1:end-60), '%f' );
            pos =  strtrim( ca2{kk}(end-60+1:end) );
            ION(:,lat2ix(lat),ut2ix(ut)) = num;
        end
    end
    end

[SO]

Sorry. I have tried this but it is not working.

[SO]

The values inside the ION array are wrong.

[per isakson]

• The value 164 is correct, I think.
• Don't be sorry, be specific! Is it systematic errors?
• You prepare some test cases.
• Now you know what kind of speed is possible with this approach.

[SO]

• Your idea is helpful. Thanks so much.
• I have uploaded another version which store all the values I want into a 3D array. The result is correct. But the speed is relatively slow than your version. (About 0.2 s, this not fast enough since I need to read thousand of these files. )
• I am not good at using 'regexp' for matching instead I am just using fgetl and strcmp.
• Any suggestion for speed improvement?

Thanks for your time!! :)

[SO]

I have made comparison with my original method.

• Searching for specific value from array is much much faster than searching in the text file each time.
• Search for 5183 values, search from array: 0.001s, search from text file each time: 260s

:)

[Stephen23]

@SO: Then maybe you should do the honorable thing and Accept per isakson's answer?

[Sadia Khaf]

Really fast getting the ION using per isakson's function cssm() but I'm having trouble with lon2ix,lat2ix,ut2ix etc. I can't seem to find any code for them in this thread. Please point me in the right direction as to what they are and where I can find them.

[Sadia Khaf]

Nevermind, found them defined inside the function cssm()

[Dogan Deniz Karadeniz]

@per isakson: Is it possible to read many files instead of your giving example (CODG1520) for your cssm function?