Re: Octave saves variable in Matlab format in large size

[David Bateman]

Muthiah Annamalai wrote:
> On Fri, 2007-02-09 at 02:26 -0500, Ge Baolai wrote:
>   
>> Hi,
>>
>> Just noticed that variables saved with '-mat-binary' option in Octave
>> result in larger file size than Matlab's. For example,
>>
>> In Matlab (ver. 7.0)
>> z = zeros(5000,999,3);
>> save 'z.mat' z
>> generates a file of size 53K bytes, which seems to contain only a
>> minimum data.
>>
>> While in Octave, the command
>> z = zeros(5000,999,3);
>> save '-mat-binary' z.mat z
>> results in a file of size 119,880,208 bytes, the full size of 5000 x 999 x 3
>> doubles.
>>
>> Any comments?
>>
>>     
> I am going to risk an explanation here. Octave is trying to be
> compatible to Matlab, might seem to be the biggest bottleneck here.
> Generally developers work in limited information, closed standards,
> and mostly sort of produce compatibility from whatever little
> documentation that exist, when nobody has time to reverse engineer the
> nuts and bolts.
>   

In fact Muthu Mathworks has fairly well documentation the file format..
I dislike their design choices of partially zipping the file rather than
just zipping the whole thing as this essentially forced me to have to
have a copy of the zipped data in memory for octave to be able to unzip.
However, the format itself is well documented, check

http://www.mathworks.com/access/helpdesk/help/pdf_doc/matlab/matfile_format.pdf

> Octave could simply use efficient storage with some/any compression
> formats (bzip, gzip etc) but then people could complain of this not
> being compatible!
>
>   
Octave does have zlib compression of its data for all formats (even
matlab's though matlab won't read it afterwards) with the "-z" flag. It
doesn't work for the HDF file formats either, as an external library is
used there...

> So its the lesser of the devils, IMHO.
>
>   
I suppose now that v7 is largely deployed it would be better to have
"-mat" and "-mat-binary" use the v7 format if that what everyone really
wants..

D.

-- 
David Bateman                                address@hidden
Motorola Labs - Paris                        +33 1 69 35 48 04 (Ph) 
Parc Les Algorithmes, Commune de St Aubin    +33 6 72 01 06 33 (Mob) 
91193 Gif-Sur-Yvette FRANCE                  +33 1 69 35 77 01 (Fax) 

The information contained in this communication has been classified as: 

[x] General Business Information 
[ ] Motorola Internal Use Only 
[ ] Motorola Confidential Proprietary