inconsistent warning for axis limits when using log scale (set_xlog/ylog) for plots with matplotlib

[DougBurke]

This is not a lien on the 4.11 release.

With the CIAO 4.11 code base (so close to master) and matplotlib version 2.2.3, I see the following behavior, where some, but not all calls to plot data/model/residuals/... leads to the following warning after calling set_xlog:

UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

Note that the presence of the warnings is inconsistent (ie I would expect it to happen all times or no times) but is consistent (in that it happens after a particular sequence of plot calls). My guess is that we are not clearing up the state from the previous plot correctly, but I haven't looked into the code.

Also note that the warning doesn't really make sense, since the data is > 0 for both axes.

% ipython --matplotlib
Python 3.5.4 (default, Sep 14 2018, 15:42:52) 
Type 'copyright', 'credits' or 'license' for more information
IPython 6.5.0 -- An enhanced Interactive Python. Type '?' for help.
Using matplotlib backend: TkAgg

In [1]: from sherpa.astro import ui

In [2]: ui.load_arrays(1, [0.2, 2, 20], [200, 20, 0.2])

In [3]: ui.set_source(ui.const1d.mdl)

In [4]: ui.plot_fit()

In [5]: ui.plot_fit_resid()

In [6]: ui.plot_fit()

In [7]: ui.plot_fit_resid()

In [8]: set_xlog()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-8-1dd939fa3894> in <module>()
----> 1 set_xlog()

NameError: name 'set_xlog' is not defined

In [9]: ui.set_xlog()

In [10]: ui.plot_fit()

In [11]: ui.plot_fit_resid()

In [12]: ui.plot_fit()
/home/djburke/testlocal/ciaot/ciao-4.11/ots/lib/python3.5/site-packages/matplotlib/axes/_base.py:3157: UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

In [13]: ui.plot_fit()

In [14]: ui.plot_fit()

In [15]: ui.plot_fit_resid()

In [16]: ui.plot_fit_resid()
/home/djburke/testlocal/ciaot/ciao-4.11/ots/lib/python3.5/site-packages/matplotlib/axes/_base.py:3157: UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

In [17]: ui.plot_fit_resid()
/home/djburke/testlocal/ciaot/ciao-4.11/ots/lib/python3.5/site-packages/matplotlib/axes/_base.py:3157: UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

In [18]: ui.plot_fit_resid()
/home/djburke/testlocal/ciaot/ciao-4.11/ots/lib/python3.5/site-packages/matplotlib/axes/_base.py:3157: UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

In [19]: ui.plot_fit()
/home/djburke/testlocal/ciaot/ciao-4.11/ots/lib/python3.5/site-packages/matplotlib/axes/_base.py:3157: UserWarning: Attempted to set non-positive xlimits for log-scale axis; invalid limits will be ignored.
  'Attempted to set non-positive xlimits for log-scale axis; '

In [20]: ui.plot_fit_resid()

In [21]: 

[DougBurke]

You can replicate this with

import numpy as np
from sherpa.data import Data1D
from sherpa import plot
from matplotlib import pyplot as plt

d = Data1D('xmpl', np.asarray([0.2, 2, 20]), np.asarray([200, 20, 0.2]))

dplot = plot.DataPlot()
dplot.prepare(d)
dplot.plot_prefs['xlog'] = True

jplot = plot.JointPlot()

jplot.plottop(dplot)
plt.clf()

If you construct a ModelPlot and send it to plottop instead then you don't see the warning. I believe it may be because here we call the matplotlib errorbar routine with both xerr and yerr as None.

[DougBurke]

Actually, it may be just xerr=None in errorbar since if you add (before the creation of dplot)

d.staterror = np.asarray([10, 1, 0.1])

then you still get the same warning (which makes sense as the message is about the x axis and it's the x axis that is tied together between the two plots).

[anetasie]

@DougBurke is this still an issue? I do not get an error in the 4.12.1 candidate release in ciao.

[DougBurke]

With the example above I still see an issue with the master branch:

%  cat p.py
import numpy as np
from sherpa.data import Data1D
from sherpa import plot
from matplotlib import pyplot as plt

d = Data1D('xmpl', np.asarray([0.2, 2, 20]), np.asarray([200, 20, 0.2]))

dplot = plot.DataPlot()
dplot.prepare(d)
dplot.plot_prefs['xlog'] = True

jplot = plot.JointPlot()

jplot.plottop(dplot)
plt.clf()

then

$ python p.py
p.py:15: UserWarning: Attempted to set non-positive left xlim on a log-scaled axis.
Invalid limit will be ignored.
  plt.clf()

[anetasie]

@DougBurke yes, I see it with this example.

[DougBurke]

I also see this with matplotlib 3.3

[DougBurke]

This appears to be due to matplotlib issue https://github.com/matplotlib/matplotlib/issues/9970 and so is not a problem with Sherpa (at least not directly)