Re: [matplotlib-devel] Polar plots

[John H. <jdh...@ac...>]

>>>>> "Roberto" == Roberto De Almeida <ral...@gm...> writes:

    Roberto> Yes, my idea is to do it "properly", but I'm still
    Roberto> getting myself familiar with the code.

    Roberto> My original plan was to create a class
    Roberto> PolarSubplot(PolarAxes), in the same way that Subplot
    Roberto> derives from Axes. I would then create the PolarAxes
    Roberto> class, implementing all the necessary methods, like
    Roberto> plot(), imshow(), etc. That's why I mentioned imshow()
    Roberto> and pcolor(), because my idea is to implement not only
    Roberto> line plots, but specially pseudo color plots on the polar
    Roberto> axes.

    Roberto> After looking at the code for transforms, though, I'm not
    Roberto> sure if all this is really necessary. It seems to me that
    Roberto> I can define a polar transform, and simply reuse all the
    Roberto> methods already defined in the Axes class to the all the
    Roberto> work, is that right? 

I believe creating a PolarAxes is the way to go.  I would probably
also derive specialized Axis classes as well to handle the r and theta
axis, using a matplotlib.patches.Circle rather than a
matplotlib.lines.Line2d for the theta axis and grid lines.  You would
probably also need to add a rotation property to the ticks class, do
that the ticks could be placed normal to the theta axis.  I could help
out here - it might call for a new line style, one along the lines of
'_', '|' (which the current ticks use) but that could be rotated.

Once you get the axis, transformations and grids set up right, yes, I
believe you will be able to reuse many of the plotting methods.  I'm
fairly certain some work will need to be done to make images, pcolors
and other plots work with though.  But the basic plot and friends
should work without modification.

    Roberto> I saw the PolarXY transform in the _transforms module,
    Roberto> but it seems to be just a stub (matplotlib 0.63.0) -- it
    Roberto> has no defined methods.

Yes and no.  The PolarXY class defines in _transforms.h

  // the api forward and inverse functions; theta in radians
  std::pair<double, double>  operator()(const double& r, const double& theta ) {
    return std::pair<double, double>( r*cos(theta), r*sin(theta) );
  }


What is missing is a NonseparableTransformation class, developed along
the lines of the SeparableTransformation, which utilizes a FuncXY
rather than funcx and funcy.  If you are not comfortable with c++ and
the pycxx extension generator package, I would be happy to add this
part.  The polar axes would set a NonseparableTransformation with the
FuncXY set to PolarXY.

    Roberto> A few more details would be great. As I said, I'm still
    Roberto> looking at the code and getting used to how things
    Roberto> work. I would be very happy to contribute with
    Roberto> matplotlib, it's a fantastic work and something that was
    Roberto> missing in the Python world for quite a long time.

Great - I've been a bit out of the loop over the last week and will be
playing catch-up this week, but will try and get the transform stuff
into CVS for you ASAP so you can work on this.

Thanks,
JDH