Re: problem plotting "N-d object"

[Ben Abbott]

On Feb 19, 2012, at 4:18 AM, Sergei Steshenko wrote:

> ----- Original Message -----
>> From: Ben Abbott <address@hidden>
>> To: CdeMills <address@hidden>
>> Cc: address@hidden
>> Sent: Sunday, February 19, 2012 1:41 AM
>> Subject: Re: problem plotting "N-d object"
>> 
>> On Feb 18, 2012, at 2:44 PM, CdeMills wrote:
>> 
>>> bpabbott wrote
>>> 
>>>> I'm not really concerned about buggy code accidentally triggering 
>> this
>>>> functionality. At the same time, if such a change is accepted, then it
>>>> needs to be documented and demos should be needed.
>>>> 
>>>> I'm not sure a new squeezable N-D array will do the job in a clean 
>> way.
>>>> Maybe I'm missing something? Are you thinking of overloading the 
>> plot
>>>> function?
>>> 
>>> Buggy code should trigger error messages ASAP. 
>>> 
>>> Now, I often run simulations and collate 1D and 2D matrices into 3D, where
>>> the third dim is the run number.  Auto-squeezing would be convenient, but I
>>> think we should have a flag to tell plot "I know my data are 3D, but 
>> please
>>> squeeze them." This way it is the INTENT to pass 3D data and have a 1D 
>> or 2D
>>> plot.
>>> 
>>> Regards
>>> 
>>> Pascal
>> 
>> Thats a good idea !
>> 
>> The plot function already accepts property name/value pairs. Is the syntax 
>> below 
>> ok ?
>> 
>>     plot (x, y, "-squeeze")
>> 
>> Since this feature may not be desired for all lines, the flag should only be 
>> applied to the line object proceeding it.
>> 
>>     plot (x1, y1, "-squeeze", x2, y2, "color", 
>> "r")
>> 
>> This would allow x1 and y1 to be squeezed (if needed) and produce a blue 
>> solid 
>> line. The second line would not be squeezed and would produce a red solid 
>> line.
>> 
>> Look ok ?
>> 
>> Ben
>> 
>> 
> 
> I think we've lost track.
> 
> To write "-squeeze" takes more effort than just "(:)".
> 
> But the root cause of my original Email is that Octave _wrongly_ thinks that 
> foo(1,1,:) is an N-d object.
> 
> The original testcase:
> 
> "
> octave:3> foo(1,1,:) = [1 2 3]
> foo =
> 
> ans(:,:,1) =  1
> ans(:,:,2) =  2
> ans(:,:,3) =  3
> 
> octave:4> plot(foo(1,1,:));
> error: transpose not defined for N-d objects
> error: called from:
> error:   
> /home/qemu/AFSWD/20111122/octave-3.4.2/share/octave/3.4.2/m/plot/private/__plt__.m
>  at line 179, column 8
> error:   
> /home/qemu/AFSWD/20111122/octave-3.4.2/share/octave/3.4.2/m/plot/private/__plt__.m
>  at line 104, column 17
> error:   
> /home/qemu/AFSWD/20111122/octave-3.4.2/share/octave/3.4.2/m/plot/plot.m at 
> line 194, column 9
> octave:4> plot(foo(1,1,:)(:));
> octave:5>             
> ".
> 
> I insist that in this case foo(1,1,:) is a _one_-d object - because two of 
> the three indexes are fixed/constant.
> 
> I.e. the plot family of functions tries to do the right thing already, but 
> because foo(1,1,:) is _not_ seen by Octave as a 1-d object the functions fail.
> 
> Regards,
>   Sergei.

The m-file language defines a 1x1xN object as an Nd array with 3 dimensions.

At the same time, the information contained in a 1x1xN array is effectively 1d.

In either event, I'd rather avoid debating semantics and prefer to focus on 
questions like ...

(1) Does implicitly squeezing the variables cause a compatibility problem today 
?
(2) Might Mathworks change the syntax for plot() in the future and break 
Octave's compatibility ?
(3) How useful is the utility of implicitly squeezing Nd arrays useful ?
(4) How harmful is the utility of implicitly squeezing Nd arrays ?

I don't see any concerns regarding (1) or (2). However, the changeset I put 
together may introduce errors for polar() and plot3(). Some testing will be 
needed to be sure bugs aren't introduced.

(3) is a matter of context. I like the idea, because I often work with matrices 
that are dependent on some variable (frequency or time for example). I use the 
3rd dimension to represent the changing dependent variable. This makes linear 
algebra convenient, since ndims(A(:,:,n)) = 2.

Regarding (4), a warning could be issued the first time the squeeze feature is 
applied.

Ben