PDL::IO::FastRaw -- A simple, fast and convenient io format for PerlDL.
This documentation refers to PDL::IO::FastRaw version 0.0.2, I guess.
use PDL; use PDL::IO::FastRaw;
writefraw($pdl,"fname"); # write a raw file
$pdl2 = readfraw("fname"); # read a raw file $pdl2 = PDL->readfraw("fname");
$pdl3 = mapfraw("fname2",{ReadOnly => 1}); # mmap a file, don't read yet
$pdl4 = maptextfraw("fname3",{...}); # map a text file into a 1-D pdl.
This is a very simple and fast io format for PerlDL. The disk data consists of two files, a header metadata file in ASCII and a binary file consisting simply of consecutive bytes, shorts or whatever.
It is hoped that this will not only make for a simple PerlDL module for saving and retrieving these files but also make it easy for other programs to use these files.
The format of the ASCII header is simply
<typeid> <ndims> <dim0> <dim1> ...
You should probably stick with the default header name. You may want
to specify your own header, however, such as when you have a large
collection of data files with identical dimensions and data types.
Under these circumstances, simply specify the Header
option in the
options hash.
The binary files are in general NOT interchangeable between different architectures since the binary file is simply dumped from the memory region of the piddle. This is what makes the approach efficient.
It is also possible to mmap the file which can give a large speedup in certain situations as well as save a lot of memory by using a disk file as virtual memory. When a file is mapped, parts of it are read only as they are accessed in the memory (or as the kernel decides: if you are reading the pages in order, it may well preread some for you).
Note that memory savings and copy-on-write are operating-system
dependent - see Core.xs and your operating system documentation
for exact semantics of whatever. Basically, if you write to a
mmapped file without ReadOnly
, the change will be reflected
in the file immediately. ReadOnly
doesn't really make it impossible
to write to the piddle but maps the memory privately so the file
will not be changed when you change the piddle. Be aware though
that mmapping a 40Mb file without ReadOnly
spends no virtual
memory but with ReadOnly
it does reserve 40Mb.
You have a whole slew of data files in ASCII from an experiment
that you ran in your lab. You're still tweaking the analysis
and plots, so you'd like if your data could load as fast as
possible. Eventually you'll read the data into your scripts
using readfraw
, but the first thing you might do is create
a script that converts all the data files to raw files:
#!/usr/bin/perl # Assumes that the data files end with a .asc or .dat extension # and saves the raw file output with a .bdat extension. # call with # >./convert_to_raw.pl file1.dat file2.dat ... # or # >./convert_to_raw.pl *.dat use PDL; use PDL::IO::FastRaw; # for saving raw files use PDL::IO::Misc; # for reading ASCII files with rcols while(shift) { # run through the entire supplied list of file names ($newName = $_) =~ s/\.(asc|dat)/.bdat/; print "Saving contents of $_ to $newName\n"; $data = rcols($_); writefraw($data, $newName); }
Now that you've gotten your data into a raw file format, you can
start working on your analysis scripts. If you scripts used rcols
in the past, the reading portion of the script should go much,
much faster now:
#!/usr/bin/perl # My plotting script. # Assume I've specified the files to plot on the command line like # >./plot_script.pl file1.bdat file2.bdat ... # or # >./plot_script.pl *.bdat use PDL; use PDL::IO::FastRaw; while(shift) { # run through the entire supplied list of file names $data = readfraw($_); my_plot_func($data); }
In the first example, I allow writefraw
to use the standard header
file name, which would be file.bdat.hdr
. However, I often measure
time series that have identical length, so all of those header files
are redundant. To fix that, I simply pass the Header option to the
writefraw
command. A modified script would look like this:
#!/usr/bin/perl # Assumes that the data files end with a .asc or .dat extension # and saves the raw file output with a .bdat extension. # call with # >./convert_to_raw.pl [-hHeaderFile] <fileglob> [-hHeaderFile] <fileglob> ... use PDL; use PDL::IO::FastRaw; # for saving raw files use PDL::IO::Misc; # for reading ASCII files with rcols my $header_file = undef; CL_OPTION: while($_ = shift @ARGV) { # run through the entire list of command-line options if(/-h(.*)/) { $header_file = $1; next CL_OPTION; } ($newName = $_) =~ s/\.(asc|dat)/.bdat/; print "Saving contents of $_ to $newName\n"; $data = rcols($_); writefraw($data, $newName, {Header => $header_file}); }
Modifying the read script is left as an exercise for the reader. :]
Sometimes you'll want to use mapfraw
rather than the read/write
functions. In fact, the original author of the module doesn't
use the read/write functions anymore, prefering to always use
mapfraw
. How would you go about doing this?
Assuming you've already saved your data into the raw format, the
only change you would have to make to the script in example 2 would
be to change the call to readfraw
to mapfraw
. That's it.
You will probably see differences in performance, though I (David
Mertens) couldn't tell you about them because I haven't played
around with mapfraw
much myself.
What if you eschew the use of writefraw
and prefer to only use
mapfraw
? How would you save your data to a raw format? In that
case, you would have to create a mapfraw
piddle with the correct
dimensions first using
$piddle_on_hd = mapfraw('fname', {Creat => 1, Dims => [dim1, dim2, ...]});
Note that you must specify the dimensions and you must tell
mapfraw
to create the new piddle for you by setting the
Creat
option to a true value, not Create
(note the missing
final 'e').
Read a raw format binary file
$pdl2 = readfraw("fname"); $pdl2 = PDL->readfraw("fname"); $pdl2 = readfraw("fname", {Header => 'headerfname'});
The readfraw
command
supports the following option:
Specify the header file name.
Write a raw format binary file
writefraw($pdl,"fname"); writefraw($pdl,"fname", {Header => 'headerfname'});
The writefraw
command
supports the following option:
Specify the header file name.
Memory map a raw format binary file (see the module docs also)
$pdl3 = mapfraw("fname2",{ReadOnly => 1});
The mapfraw
command
supports the following options (not all combinations make sense):
If creating a new file or if you want to specify your own header data for the file, you can give an array reference and a scalar, respectively.
Create the file. Also writes out a header for the file.
Set the file size. Automatically enabled with Creat
. NOTE: This also
clears the file to all zeroes.
Disallow writing to the file.
Specify the header file name.
Memory map a text file (see the module docs also).
Note that this function maps the raw format so if you are using an operating system which does strange things to e.g. line delimiters upon reading a text file, you get the raw (binary) representation.
The file doesn't really need to be text but it is just mapped as one large binary chunk.
This function is just a convenience wrapper which firsts stat
s
the file and sets the dimensions and datatype.
$pdl4 = maptextfraw("fname", {options}
The options other than Dims, Datatype of mapfraw
are
supported.
Should be documented better. writefraw
and readfraw
should
also have options (the author nowadays only uses mapfraw
;)
Copyright (C) Tuomas J. Lukka 1997. All rights reserved. There is no warranty. You are allowed to redistribute this software / documentation under certain conditions. For details, see the file COPYING in the PDL distribution. If this file is separated from the PDL distribution, the copyright notice should be included in the file.