konvert

Konvert implements universal conversion graphs for e.g. coordinate transformations.

At the core, konvert represents types connected by conversions on a graph. Such conversion graphs can be expandend incrementally, because it only requires two connections to connect a new type to the graph, This type can then be freely converted to any other type in the graph, by automatically chaining conversions between existing types.

The automatic chaining requires the conversions to be parameter-free. Conversions with additional parameters are called projections, which konvert also helps you manage.

Points

The moule konvert.points includes a conversion graph for coordinate transformations. As an example consider a set of points in two-dimensional Cartesian coordinates:

from konvert.points import Cartesian2D

line = Cartesian2D(x=[0, 1, 2], y=[0, 1, 2])

This points can be converted to other implented 2D point sets, like Polar and Bipolar, by using the to() method

from konvert.points import Polar, Bipolar

lp = line.to(Polar)
lb = line.to(Bipolar)

It can also be lifted up to its 3D description, l3 = line.to(Cartesian3D). The conversion graph can automatically chain conversions, so in fact, any point in the plane can be hoisted up to any point in three-dimensional space. That means, that points in e.g. polar coordinates, can be hoisted up to Cartesian3D.

p0 = Polar(theta=30 * degrees, r=1.5).to(Cartesian3D)

Points in 3D can be represented using the coordinate representations Cartesian3D, Cylindrical, and Spherical. All Cartesian types have some helper methods for easily manipulating points.

p0 = Cartesian3D(1, 1, 1)
q0 = Cartesian3D(1, 1, 0)

# Create normalized version
p1 = p0.normalized()

# Shift p0 by q0
p0.shift(q0)  

# Rotate around an axis thorugh q0.
p0.rotate(theta=45 * degrees, point=q0)

Helpers

Because the points module work extensively with angles, konvert defines a convenience degrees symbol, which converts values and arrays in degrees to radians,

from konvert.points import degrees

theta = 90 * degrees
# theta is now pi/2

theta = numpy.array([45, 60, 90, 180, 360]) * degrees
# theta is now numpy.array([pi/4, pi/3, pi/2, pi, 2*pi])

Map projections

The points module also contains a set of map projections: Azimuthal, AzimuthalEquidistant, Mercator, Orthographic, Stereographic and Equirectangular. Points are represented on the Sphere, in spherical coordinates, or in Equitorial coordinates (lattitude and longitude).

points = Equitorial([40.12, 50.53] * degrees, [33.16, 44.53] * degrees, r=1)

p0 = points.project(Mercator)
p1 = points.project(Stereographic)
...

The points can be easily plotted using matplotlib and the plot utility on Cartesian2D, p0.to(Cartesian2D).plot().

Extending the conversion graph

It is fairly simple to extend an existing conversion graph and an existing projection collection. Let us extend the points graph with a Skew2D coordinate system. Note the use of numpy to efficiently represent and transform coordinates.

import numpy
from konvert.points import Points

class Skew2D(Points):
    _sig = ('x', 'y', 'theta')

    def __init__(self, x, y, theta):
        """ 
        Skew 2D coordinates. Theta is a scalar denominating the skew angle.
        """
        self.x = numpy.array(x)
        self.y = numpy.array(y)
        self.theta = theta

The Skew2D class can be connected to the konvert.points conversion graph by using the following conversion.

from konvert.points import converters, Conversion

@converters.register()
class Skew2DToCartesian2D(Conversion):
    src = Skew2D
    dst = points.Cartesian2D

    @staticmethod
    def convert(skew):
        return points.Cartesian2D(skew.x + numpy.cos(theta) * skew.y, numpy.sin(theta) * skew.y)

The reverse conversion requires the additional theta parameter and must be implemented as a projection.

from konvert.points import projectors, Projection

@projectors.register()
class Cartesian2DToSkew2D(Projection):
    src = points.Cartesian2D
    dst = Skew2D

    @staticmethod
    def project(cart, theta=np.pi / 2):
        y = cart.y / numpy.sin(theta)
        x = cart.x - numpy.cos(theta) * y
        return Skew2D(x, y, theta)

With these two additions we can convet between Skew2D points and any points type in the graph,

p0 = Skew2D(1, 2, theta=45 * degrees)
p1 = p0.to(Cylindrical)

or the opposite way,

p1 = Cylindrical(theta=30 * degrees, phi=45 * degrees, r=10)
p0 = p1.project(OnPlane).project(Cartesian2DToSkew2D, theta=45 * degrees)

Extensions

The conversion graph has been implemented in the conversions module. It is possible to create additional conversion graphs and register existing or new conversions in those graphs. As an example we may create a colors module, which converts between values in the RGB and HSL color space. For this is example we will not bother with vectorizing the entities, and instead just work with single color entries and use the colorsys module in the python standard library.

import colorsys
from konvert.conversions import Conversion, ConversionGraph

converters = ConversionGraph()

class Color(metaclass=ABCMeta):     
     def to(self, type):
        return convertes.convert(self, type)


class RGB(Color):
    def __init__(self, r, g, b):
        self.r = r
        self.g = g
        self.b = b

class HSL(Color):
    def __init__(self, h, s, l):
        self.h = h
        self.s = s
        self.l = l

converters.register()
class RGBToHSL(Conversion):
    src = RGB
    dst = HSL

    @staticmethod
    def convert(rgb):
        hsl = colorsys.rgb_to_hsl(rgb.r, rgb.g, rgb.b)
        return HSL(*hsl)


converters.register()
class HSLToRGB(Conversion):
  src = RGB
  dst = HSL

  @staticmethod
  def convert(hsl):
      rgb = colorsys.hsl_to_rgb(hsl.h, hsl.s, hsl.l)
      return RGB(*rgb)

We can now write RGB(0.1, 0.3, 0.3).to(HSL) and get the correct result.

If we wanted, we could merge this graph into the points conversion graph, or include part of the points graph in the color conversion graph...