MathGroup Archive 2003

[Date Index] [Thread Index] [Author Index]

Search the Archive

RE: IO-Performance of Mathematica 4.1

  • To: mathgroup at smc.vnet.net
  • Subject: [mg44743] RE: [mg44684] IO-Performance of Mathematica 4.1
  • From: "Benedetto Bongiorno" <BBongiorno at attglobal.net>
  • Date: Mon, 24 Nov 2003 00:05:29 -0500 (EST)
  • Reply-to: <BBongiorno at attglobal.net>
  • Sender: owner-wri-mathgroup at wolfram.com

I too have speed problems with importing large data sets. After many years
of experimenting with various methods, I have concluded:
1 - Import & Export commands work best on *.csv 2 dimensional matrix files.
2 - Mathematica is not a speed demon on I/O functions.



-----Original Message-----
From: Marko Kastens [mailto:Kastens at Hamburg.baw.de]
To: mathgroup at smc.vnet.net
Subject: [mg44743] [mg44684] IO-Performance of Mathematica 4.1


Hi!

I'am searching for a good (fast) way to import big datasets in Mathematica
4.1.

What I've done so far:
- The read-in from the _orignal_ (ASCII)file needs 97 sec. (a 5xxxx x 2
matrix)
- Exporting my matrix using Export[] with "List" or "Table" needs
(nearly) infinity of time :-(
- Exporting the matrix with the HDF (Hierarchy Data Format): 7 seconds
:-) Great, but...
- Importing the matrix with HDF: fast but wrong: the HDF seems to
supports only Real64. My first column is a 10-digit-Integer representing
the MM-time-format. By converting this to yyExx the last digits get lost
and my time is doing funny jumps...

- using the Experimental-Functions like this:

<< Experimental`
t0 = AbsoluteTime[];
strm = OpenWrite["tmp0.bin", DOSTextFormat -> False];
BinaryExport[strm, Length[data], "Integer32"]
For[i = 1, i < Length[data] + 1,
   {
      BinaryExport[strm, data[[i]], {"Integer64", "Real64", "Integer64"}]
   }; i++]
Close[strm];
Print[AbsoluteTime[] - t0];

it takes 415 seconds to export and 247 seconds to import. Even worser
than importing the original.

Only the good old ASCII-way is better:

t0 = AbsoluteTime[];
strm = OpenWrite["tmp1.dat"];
Write[strm, Length[data]]
For[i = 1, i < Length[data] + 1,
   {
     Write[strm, data[[i, 1]]],
     Write[strm, data[[i, 2]]],
     Write[strm, data[[i, 3]]]
     }; i++]
Close[strm];
Print[AbsoluteTime[] - t0];

It takes 32 seconds to export and 16 seconds to import.


Hmm, normaly binary-import/export is faster than ascii-import/export.
Does anybody know a proper and fast method? Thanks a lot.

marko


  • Prev by Date: Re: Calling Mathematica via MathLink multiple times
  • Next by Date: RE: Mesh generation in 2D
  • Previous by thread: IO-Performance of Mathematica 4.1
  • Next by thread: Re: IO-Performance of Mathematica 4.1