Description

astrofrog

opened on May 17, 2011

Issue body actions

The following code should produce an image that is 3600x3216 and completely blue, but there is a 1-pixel white border on the bottom and left:

import numpy as np

import matplotlib
matplotlib.use('Agg')
import matplotlib.pyplot as mpl

fig = mpl.figure(figsize=(12.00, 12.00 * 3216. / 3600.))
ax = fig.add_axes([0.,0.,1.,1.])

ax.set_autoscale_on(False)

ax.xaxis.set_view_interval(0.5, 3600.5)
ax.yaxis.set_view_interval(0.5, 3216.5)

for tick in ax.xaxis.get_ticklines():
    tick.set_visible(False)

for tick in ax.yaxis.get_ticklines():
    tick.set_visible(False)

for label in ax.xaxis.get_ticklabels():
    label.set_visible(False)

for label in ax.yaxis.get_ticklabels():
    label.set_visible(False)

for key in ax.spines:
    ax.spines[key].set_visible(False)

ax.imshow(np.zeros((3216, 3600)), extent=(0.5, 3600.5, 0.5, 3216.5), origin='lower')
fig.savefig('test.png', dpi=300, transparent=False)

If autoscale is turned on, it looks fine, but I need to keep autoscale off for other reasons, so I want to understand what's going on here. If I do:

print ax.xaxis.get_view_interval()
print ax.yaxis.get_view_interval()

I get:

[    0.   3600.5]
[    0.   3216.5]

which must be part of the issue here, since the minima should be 0.5, not 0. If I use set_xlim and set_ylim instead of ax.xaxis.set_view_interval, it works fine, so there must be a bug in set_view_interval?

Note that this issue is not related to #117, and occurs even if the patch in #117 is applied.

Activity

1. astrofrog commented on May 21, 2011

   

   astrofrog

   on May 21, 2011

   ContributorAuthor

   More actions

   In def set_view_interval in axis.py, should the following lines:

           Vmin, Vmax = self.get_view_interval()
           if Vmin < Vmax:
               self.axes.viewLim.intervalx = (min(vmin, vmax, Vmin),
                                              max(vmin, vmax, Vmax))
           else:
               self.axes.viewLim.intervalx = (max(vmin, vmax, Vmin),
                                              min(vmin, vmax, Vmax))
   

   Not be the following?

           if Vmin < Vmax:
               self.axes.viewLim.intervalx = (min(vmin, vmax),
                                              max(vmin, vmax))
           else:
               self.axes.viewLim.intervalx = (max(vmin, vmax),
                                              min(vmin, vmax))
   

   Basically, the old limits should be used to determine the order, but apart from that, why should they appear in the min(...) and max(...) expressions?
2. leejjoon commented on May 24, 2011

   

   leejjoon

   on May 24, 2011

   Contributor

   More actions

   The changelog entry of the associated commit seems to indicate that this is an intended behavior.

   cb5cab4

   @efiring, can you comment on this?

   For example,

    ax = plt.subplot(111)
   
    ax.xaxis.set_view_interval(5, 10, ignore=False)
    print ax.xaxis.get_view_interval()
   
    ax.xaxis.set_view_interval(5, 10, ignore=True)
    print ax.xaxis.get_view_interval()

   prints

   [  0.  10.]
   [  5.  10.]

   which seems a bit inconsistent.
3. efiring commented on May 24, 2011

   

   efiring

   on May 24, 2011

   Member

   More actions

   On 05/23/2011 06:57 PM, leejjoon wrote:

   The changelog entry of the associated commit seems to indicate that this is an intended behavior.

   cb5cab4

   @efiring, can you comment on this?

   The problem here is that Axis.set_view_interval should be an obscure
   private method, or it should be broken up into two methods, perhaps one
   or both of which should also be obscure and private; its only internal
   use, as far as I can see, is in ticker.py. The methods that are
   intended for public use in setting the viewLim are the Axes methods.
   The original implementation of Axis.set_view_interval was broken for its
   original purpose; I fixed that, but did not put enough additional
   explanation in the docstring (which was previously missing entirely, and
   maybe should have been left so) to fully explain what the ignore kwarg
   does--which is to change the behavior so that it not only preserves the
   original order (hence axis inversion, if any), but also permits only
   expansion, not contraction, of the interval. It is this very different
   sort of functionality, required by ticker, that is causing the confusion.

   So: minimal solution is to improve the docstring; a cleaner long-term
   solution would involve ending up with two methods instead of one, and no
   kwarg. How to do this in with minimal user code breakage and with good,
   readable names for all methods is not obvious to me yet.

   Eric

   For example,

     ax = plt.subplot(111)
   
     ax.xaxis.set_view_interval(5, 10, ignore=False)
     print ax.xaxis.get_view_interval()
   
     ax.xaxis.set_view_interval(5, 10, ignore=True)
     print ax.xaxis.get_view_interval()

   prints

   [  0.  10.]
   [  5.  10.]

   which seems a bit inconsistent.
4. astrofrog commented on May 24, 2011

   

   astrofrog

   on May 24, 2011

   ContributorAuthor

   More actions

   Ok, I understand now that these methods are not meant to be public. I'm not sure why I originally used them instead of set_xlim and set_ylim, but I must have stumbled upon them and assumed they were public. I'm just going to switch to set_xlim and set_ylim for my code.
5. ghost assigned

   efiring

   on Jun 15, 2011
6. efiring

   added a commit that references this issue on Aug 20, 2011

   axis.py: clarify docstring for set_view_interval. Closes issue #118.

   a96422e
7. efiring

   closed this as completedon Aug 20, 2011

Issue actions