Re: [Matplotlib-users] problem with x scale on error bar plot

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

>>>>> "Fl=E1vio" =3D=3D Fl=E1vio Code=E7o Coelho <fcc...@fi...> wr=
ites:

    Fl=E1vio> Hi john, I am getting a bad autoscale when I generate a
    Fl=E1vio> two point y-error errorbar plot:

    Fl=E1vio> As you can see in the figure attached the points are
    Fl=E1vio> falling on the margin of the plot.

    Fl=E1vio> I think a space before the first point and after the last
    Fl=E1vio> point of the plot should be added automatically, since no
    Fl=E1vio> one will want a plot like this by default.

This is a fix that is trivial to implement.  But I would like to get
some feedback because changing the default would affect every almost
all matplotlib plots.

The question is: should the autoscale endpoints always exceed the data
end points, or are there some cases in which it is desirable for the
datalim and viewlim to coincide?  In this case

    from matplotlib.matlab import *
    plot([1,2], 'o')
    show()

it seems like you want the viewlim to exceed the datalim.  In cases
like

    from matplotlib.matlab import *
    t =3D linspace(0,1,100)
    plot(t, sin(2*pi*t))
    show()

it looks to me like having the xlim at 0,1 is the best solution.

To change the behavior, edit matplotlib/ticker.py and search for class
MultipleLocator.  In the autoscale method, change

        vmin =3D self.base.le(dmin)
        vmax =3D self.base.ge(dmax)

to
        vmin =3D self.base.lt(dmin)
        vmax =3D self.base.gt(dmax)

and try both of the examples above in the le/ge case and in the lt/gt
case.

JDH