Dear pyx experts:

I'm curious if there is a way
> to specify the y2 axis as a function of the y axis? I am particularly
> interested in cases where the axes are not linearly related but are
> nonetheless monotonic. In the case that I am currently working on,
> the axes are related by y2 = y/(y+a) where a is a constant. Please
> let me know if this may be possible at any level of (in)elegance.

On 08.07.05, Andre Wobst wrote:

On 08.07.05, Tom Kornack wrote:

> I often use y and y2 axes on my graphs. I'm curious if there is a way
> to specify the y2 axis as a function of the y axis? I am particularly
> interested in cases where the axes are not linearly related but are
> nonetheless monotonic. In the case that I am currently working on,
> the axes are related by y2 = y/(y+a) where a is a constant. Please
> let me know if this may be possible at any level of (in)elegance.

Well, its a quite complicated problem as soon as you do not fix the
axis ranges. The problem is, that for unfixed axis ranges, the
partitioning can have some influence on the axis range. So to make a
simple solution, we fix the axis ranges and go for it. Than you can do
the following (on top of PyX 0.7.1):

from pyx import *

class relatedaxis(object, graph.axis.lin):
def __init__(self, otheraxis, a, **kwargs):
self.otheraxis = otheraxis
self.a = a
graph.axis.lin.__init__(self, **kwargs)
def convert(self, value):
return self.otheraxis.convert(self.a*float(value)/(1.0-float(value)))
def invert(self, value):
otherinvert = self.otheraxis.invert(value)
return otherinvert/(otherinvert + self.a)
min = property(lambda self: self.invert(0), lambda self, min: None)
max = property(lambda self: self.invert(1), lambda self, max: None)

yaxis = graph.axis.lin(min=0, max=1)
y2axis = relatedaxis(yaxis, 1)
g = graph.graphxy(width=8, y=yaxis, y2=y2axis, x=graph.axis.lin(min=0, max=0.5))
g.plot(graph.data.function("y2=x"))
g.writeEPSfile("relatedaxis")

The y2 axis is full functional except that its ranges do completely
depend on the y axis. Another difficult thing is to get a proper axis
partitioning for this case. You can use the autolinear-parter (i.e.
the default), but I'm not sure whether equal distant ticks (in
y2-values) are the right thing for you. Unfortunately you can't use
the autologarithmic-parter (not just because we have a zero at the
axis in our example above, but also because the convert method always
fails for value=1). However, you can always place manual ticks and set
parter to None, or you can write your own (automatic) parter.