Re: Non-linear axis data in imagesc

[Ben Abbott]

On Dec 2, 2011, at 9:12 AM, Daniel Arteaga wrote:

> On 02/12/11 14:29, Ben Abbott wrote:
>> 
>> On Dec 2, 2011, at 8:20 AM, Daniel Arteaga wrote:
>> 
>>> On 02/12/11 14:15, Ben Abbott wrote:
>>>> On Dec 2, 2011, at 6:52 AM, Daniel Arteaga wrote:
>>>> 
>>>>> Hi all,
>>>>> 
>>>>> imagesc fails to display correctly non-linarly spaced axes. Instead it 
>>>>> "invents" linear axis from the maximum value. See what I mean:
>>>>> 
>>>>> x = logspace(0,5,20);
>>>>> y = logspace(0,2,10);
>>>>> A = rand(10,20);
>>>>> imagesc(x,y,A);
>>>>> 
>>>>> Is this a bug? Any workaround to get the correct behaviour?
>>>>> 
>>>>> Tested using GNU Octave Version 3.4.2 under Ubuntu 11.10.
>>>>> 
>>>>> Thank you,
>>>>> 
>>>>> Daniel
>>>> 
>>>> If you see (0,0) in the upper-left, then it is working as intended. Matlab 
>>>> behaves this way as well.
>>> 
>>> Sorry, I was not meaning that. What I was meaning is that the axis labels 
>>> do not correspond to the x vector defined above. The real axis should be 
>>> logarithmic, but imagesc invents a linear axis.
>>> 
>>> Thank you,
>>> 
>>> Daniel
>> 
>> Ok. That *feature* is also inherited from Matlab.
>> 
>> I've attached the result from Matlab (R2011b). Does it look the same as what 
>> Octave produces for you?
> 
> Yes, the result is essentially the same.
> 
> Anyway, be in Matlab or Octave, the result is incorrect as I understand, 
> since it is assigning values to the axis that do not match with the plotted 
> data.
> 
> In other words, for a given x and y value (according to the displayed axis), 
> the color plotted is not the correct one.
> 
> In my opinion, at least a warning should be displayed in this situation.
> 
> Thank you very much,
> 
> Daniel

The ML implementation is ugly. Adding a warning is a good idea.

I'll apply a patch.

Thanks,
Ben