Re: how to handle negative data and position values for log scales

[Ben Abbott]

On Nov 9, 2010, at 4:17 AM, Rik wrote:

>> Should the change I just pushed be left in place, or shall negative values 
>> for different objects be handled differently for each object type for log 
>> scale plots?
>> 
>> For example, what if text objects for a semilogy plot have a negative 
>> y-positions? Perhaps my change should remain and something special be done 
>> for text objects?
>> 
>> Ben
>> 
> Ben,
> 
> I think your patch is fine.  Although it might be clearer to move the code
> into each of the sub-objects which is affected, it would also create a lot
> of redundancy.  When we next need to tweak this issue we'd find ourselves
> hunting through the code to find all the occurrences to update.  Your
> current code is concise and represents just one instance to update.
> 
> As for negative y-positions, there is currently no support for it at the
> axis level.  That would need to be addressed before considering whether
> objects can be placed at negative locations.  In my opinion this is too
> small a corner case to worry about.  If you're interested though, try
> running a semilog plot on Matlab and forcing the y-axis to display negative
> values.  What do they do?
> 
> Note that confining placement to the visible axis isn't really much of a
> restriction.  For example, the left-axis on the plotyy demo goes to -1, but
> the right-axis is log scale and starts at 10^-1.  One can still place
> objects that seem to occur at negative positions relative to the left-axis
> by using the second axis and y-positions that are positive and above 10^-1.
> 
> Cheers,
> Rik

I'll favor avoiding redundancy, and leave the patch as it is.

I'll also place the text problem on my list of things to look at.

Ben