Re: save function

[Mike Miller]

On Tue, 8 Mar 2005, John W. Eaton wrote:

> On  6-Mar-2005, Mike Miller <address@hidden> wrote:
>
> | That is a reason to have the feature, but it is not a reason to impose it
> | on all users for every data file they save.  We often create data files in
> | octave to be imported into other programs.  It would be very nice if we
> | could save in ascii format without the header lines.  Something like this
> | would do:
> |
> |    `-ascii2'
> |          save the data as plain text without Octave header information
> |
> | Haven't we had a lot of questions/complaints that would be resolved by the
> | addition of this feature?
> |
> | Maybe this would work only for 2-d arrays (matrices), but that is often
> | the sort of thing we are trying to save:  Table data where rows are
> | records and columns are fields.  Octave save -ascii does what we want
> | except that it has those extra initial lines that must be removed by some
> | external program before we can use the data.
>
> Sorry, I didn't see this message before I sent my previous reply.
>
> I don't think we want -ascii2 to mean "do the Matlab-compatible -ascii 
> thing" because it just trades one incompatibility for another.  If we 
> want compatibility, then we have to make Octave's "save -ascii" do what 
> Matlab's does.  That includes saving character strings as numeric data 
> and allowing multiple values to be stored in a single file (with no 
> header information that gives you a clue about what variables are where 
> -- I guess when saving the entire workspace, the variables are stored in 
> alphabetical order?) and so on.  We can still keep Octave's old -ascii 
> behavior, but we will need rename the option.  As I noted in my previous 
> message, this is likely to cause some pain.  How can we minimize it?


I guess we need to hear from some of the people who would find the change 
painful.  It wouldn't affect me much.

Regarding MATLAB compatibility - I think MATLAB's "save -ascii" behavior 
is not so nice.  I wonder if people really use it to save multiple 
variables in a single file.  I also don't like the way MATLAB stores 
values like this...

   1.0000000e+00   2.0000000e+00   3.0000000e+00

...instead of like this:

 1 2 3

The former wastes a lot of space, but the MATLAB idea is to align output 
in columns like this:

  1.0000000e+000             Inf             NaN  1.3700000e+201
  8.1797434e+006  3.4197062e+006
  6.6022756e+006  2.8972590e+006
 -3.0461737e+006 -3.0276440e+006 -3.7837300e+006 -4.9655245e+006
 -1.8965375e+006 -5.4167385e+006 -8.6001160e+006 -8.9976918e+006
 -1.9343116e+006 -1.5087298e+006 -8.5365513e+006 -8.2162916e+006
 -6.8222322e+006 -6.9789848e+006 -5.9356291e+006 -6.4491038e+006
  9.8000000e+001  1.1100000e+002  9.8000000e+001
  1.0600000e+002  1.1100000e+002  1.0100000e+002
  1.1600000e+002  1.1100000e+002  1.0900000e+002
  1.0600000e+002  9.7000000e+001  1.1000000e+002

That is MATLAB "save -ascii" output for four variables.  The first is 1x4, 
the second is 2x2, the third is 4x4 and the last is a 4x3 string matrix:

x=[1 Inf NaN 13.7*10^200];
x2=10000000*rand(2);
x3=-10000000*rand(4);
x4=['bob';'joe';'tom';'jan'];
save -ascii c:\x4.txt x x2 x3 x4

The translation of strings to numbers is probably not what people want, 
but it gets worse.  Here's how MATLAB (v 5.2 anyway) handles complex 
numbers:

# e=eig(rand(3))

e =

   1.4768
   0.1346 + 0.1774i
   0.1346 - 0.1774i

# save -ascii c:\e.txt e

This is all it saves:

  1.4768417e+000
  1.3460833e-001
  1.3460833e-001

MATLAB's "save -ascii" simply deletes the imaginary part of the complex 
number!

So, I think we'd better proceed very carefully toward compatibility in 
this case.  It's easy to see why you (John) didn't make Octave's "save 
-ascii" compatible with MATLAB's in the first place.

Maybe we should be using "save -matlab -ascii" to turn on MATLAB 
compatibility in this case because it is so badly broken.

Regarding my idea to add a -ascii2 option:  I think maybe -ascii_noheaders 
would be a better name, and I would use that, but I have no use for a 
fully MATLAB compatible "save -ascii" now that I know what it does!

Mike

--
Michael B. Miller, Ph.D.
Assistant Professor
Division of Epidemiology and Community Health
and Institute of Human Genetics
University of Minnesota
http://taxa.epi.umn.edu/~mbmiller/



-------------------------------------------------------------
Octave is freely available under the terms of the GNU GPL.

Octave's home on the web:  http://www.octave.org
How to fund new projects:  http://www.octave.org/funding.html
Subscription information:  http://www.octave.org/archive.html
-------------------------------------------------------------