Textscan vs importdata

Hi,

I though textscan was faster than importdata but it seems to me that for bigger file importdata is faster. Any explanations?

Here is the code :

tic
  raw1=importdata('185mo.csv');
toc 
  tic
    fid = fopen('185mo.csv');
    raw2 = textscan(fid, '%f %f %f', 'delimiter', ',', ...
                     'EmptyValue', -NaN,'HeaderLines',16);
    fclose(fid);
  toc

Here is the result :

Elapsed time is 30.462402 seconds.
Elapsed time is 38.665811 seconds.

[Oleg Komarov]

Consider that importdata uses textscan. I would suggest to go in debug mode to check what options they use, but I suspect it's the NaN padding.

[Walter Roberson]

Interesting, I did not know that -NaN had a different representation than NaN, but it seems to (in R2008b at least)

>> sprintf('%16lx\n',typecast(nan,'uint64'))

ans =

fff8000000000000

>> sprintf('%16lx\n',typecast(-nan,'uint64'))

ans =

7ff8000000000000

[Benoit]

Well I've tried without EmptyValue,

tic

raw1=importdata('185mo.csv');

toc

tic

fid = fopen('185mo.csv');

raw2 = textscan(fid, '%f %f %f', 'delimiter', ',', ...

'HeaderLines',16);

fclose(fid);

toc

Here is the result:

Elapsed time is 28.735443 seconds.

Elapsed time is 36.212666 seconds.

[Jan]

@Walter: In R2009a NaN and -NaN are "000000f8ff" on a Win7/64 PC.

Answer 1

[xiangsheng guan]

I find that importdata uses fread for .txt files. But I can access neither source file of fread nor textscan.