New contourf() drawing polygon boundaries for some reason?

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

···

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

So you see this behavior if you switch to the Agg backend? That's the backend used to generate the images in the gallery. If there's a difference there, that would seem to suggest some tweaking of the macosx backend (which is still relatively new) is in order.

Mike

Zane Selvans wrote:

···

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

Um, yeah. So my response got bounced because of the attachment. Take 2:

For some reason my script bombed when I switched to the Agg backend,
trying to display to the screen (it said Figure has no method show())

So I output the plot as both a PDF and a PNG (still having backend:
agg in my rcfile) and in both of those cases, irregular gaps are
visible between the polygons making up the filled contours. This
wasn't the case with my previously installed setup. It looks as if
for some reason the vertices of the filled polygons are being
calculated differently from different sides of the same contour,
leading to overlap in some places, and gaps in others. You can download
the PDF version (in which the exact geometry is much clearer).
from:

http://zaneselvans.org/dropbox/LinDensity_Grid.pdf

Zane

···

On Fri, Jun 12, 2009 at 5:51 AM, Michael Droettboom<mdroe@...86...> wrote:

So you see this behavior if you switch to the Agg backend? That's the
backend used to generate the images in the gallery. If there's a difference
there, that would seem to suggest some tweaking of the macosx backend (which
is still relatively new) is in order.

Mike

Zane Selvans wrote:

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

Shot in the dark here, but what if you set the rcParam "path.simplify" to False? There have been recent changes to that code.

Also, since the Agg backend doesn't have an associated GUI, you need to use the savefig() command and provide a filename, rather than using show().

Cheers,
Mike

Zane Selvans wrote:

···

Um, yeah. So my response got bounced because of the attachment. Take 2:

For some reason my script bombed when I switched to the Agg backend,
trying to display to the screen (it said Figure has no method show())

So I output the plot as both a PDF and a PNG (still having backend:
agg in my rcfile) and in both of those cases, irregular gaps are
visible between the polygons making up the filled contours. This
wasn't the case with my previously installed setup. It looks as if
for some reason the vertices of the filled polygons are being
calculated differently from different sides of the same contour,
leading to overlap in some places, and gaps in others. You can download
the PDF version (in which the exact geometry is much clearer).
from:

http://zaneselvans.org/dropbox/LinDensity_Grid.pdf

Zane

On Fri, Jun 12, 2009 at 5:51 AM, Michael Droettboom<mdroe@...86...> wrote:
  

So you see this behavior if you switch to the Agg backend? That's the
backend used to generate the images in the gallery. If there's a difference
there, that would seem to suggest some tweaking of the macosx backend (which
is still relatively new) is in order.

Mike

Zane Selvans wrote:
    

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

If I set path.simplify: False, the shape of the gaps between the
filled polygons does change. Instead of being irregular, it becomes
an infinitessimally thin gap of uniform width, allowing the (in this
case white) background to show through.

In both of these cases (path.simplify: True|False), the PNG version of
the same figures also show representations of these gaps which are
identical to those which appear in the PDF (though obviously
pixelated), so I don't think it's something that's wrong in the vector
graphics code per se.

Zane

···

On Fri, Jun 12, 2009 at 11:46 AM, Michael Droettboom<mdroe@...86...> wrote:

Shot in the dark here, but what if you set the rcParam "path.simplify" to
False? There have been recent changes to that code.

Also, since the Agg backend doesn't have an associated GUI, you need to use
the savefig() command and provide a filename, rather than using show().

Cheers,
Mike

Zane Selvans wrote:

Um, yeah. So my response got bounced because of the attachment. Take 2:

For some reason my script bombed when I switched to the Agg backend,
trying to display to the screen (it said Figure has no method show())

So I output the plot as both a PDF and a PNG (still having backend:
agg in my rcfile) and in both of those cases, irregular gaps are
visible between the polygons making up the filled contours. This
wasn't the case with my previously installed setup. It looks as if
for some reason the vertices of the filled polygons are being
calculated differently from different sides of the same contour,
leading to overlap in some places, and gaps in others. You can download
the PDF version (in which the exact geometry is much clearer).
from:

http://zaneselvans.org/dropbox/LinDensity_Grid.pdf

Zane

On Fri, Jun 12, 2009 at 5:51 AM, Michael Droettboom<mdroe@...86...> >> wrote:

So you see this behavior if you switch to the Agg backend? That's the
backend used to generate the images in the gallery. If there's a
difference
there, that would seem to suggest some tweaking of the macosx backend
(which
is still relatively new) is in order.

Mike

Zane Selvans wrote:

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

I switched back to using the macosx backend, and it turns out that the
thin black lines surrounding the polygons (including crossing the
filled contour regions from one closed contour to another) only get
displayed in the GUI. PDF and PNG output look fine.

Zane

···

On Fri, Jun 12, 2009 at 2:27 PM, Zane Selvans<zane@...1923...> wrote:

If I set path.simplify: False, the shape of the gaps between the
filled polygons does change. Instead of being irregular, it becomes
an infinitessimally thin gap of uniform width, allowing the (in this
case white) background to show through.

In both of these cases (path.simplify: True|False), the PNG version of
the same figures also show representations of these gaps which are
identical to those which appear in the PDF (though obviously
pixelated), so I don't think it's something that's wrong in the vector
graphics code per se.

Zane

On Fri, Jun 12, 2009 at 11:46 AM, Michael Droettboom<mdroe@...86...> wrote:

Shot in the dark here, but what if you set the rcParam "path.simplify" to
False? There have been recent changes to that code.

Also, since the Agg backend doesn't have an associated GUI, you need to use
the savefig() command and provide a filename, rather than using show().

Cheers,
Mike

Zane Selvans wrote:

Um, yeah. So my response got bounced because of the attachment. Take 2:

For some reason my script bombed when I switched to the Agg backend,
trying to display to the screen (it said Figure has no method show())

So I output the plot as both a PDF and a PNG (still having backend:
agg in my rcfile) and in both of those cases, irregular gaps are
visible between the polygons making up the filled contours. This
wasn't the case with my previously installed setup. It looks as if
for some reason the vertices of the filled polygons are being
calculated differently from different sides of the same contour,
leading to overlap in some places, and gaps in others. You can download
the PDF version (in which the exact geometry is much clearer).
from:

http://zaneselvans.org/dropbox/LinDensity_Grid.pdf

Zane

On Fri, Jun 12, 2009 at 5:51 AM, Michael Droettboom<mdroe@...86...> >>> wrote:

So you see this behavior if you switch to the Agg backend? That's the
backend used to generate the images in the gallery. If there's a
difference
there, that would seem to suggest some tweaking of the macosx backend
(which
is still relatively new) is in order.

Mike

Zane Selvans wrote:

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

If I set path.simplify: False, the shape of the gaps between the
filled polygons does change. Instead of being irregular, it becomes
an infinitessimally thin gap of uniform width, allowing the (in this
case white) background to show through.

Just to clarify, these backs only show with the Mac OS-X gui window,
not with the PNG or PDF output (which are actually handled by the Agg
and PDF backends respectively, and don’t use the Cocoa/Quartz-specific
code in the macosx backend).

If that’s the case, I’ll forward this thread to the attention of
Michiel de Hoon, the author of the macosx backend.

As for path.simplify = True causing irregular gaps, that’s my area, and
I’ll look into that further. path.simplify should not be creating
visible artifacts (at least in raster images such as PNG), so that’s a
bug.

Cheers,

Mike

···

<mdroe@…86…>http://zaneselvans.org/dropbox/LinDensity_Grid.pdf<mdroe@…86…>http://macinscience.org/?page_id=6

Yes. By far the worst of these behaviors is the macosx GUI output. I
could see the other ones just being the way it's supposed to look.
Here's a summary:

backend: macosx; path.simplify: (false|true) GUI => black borders to
drawn polygons (incl. contour region crossing lines, very bad)
http://zaneselvans.org/dropbox/contourf_backendmacosx.png

backend: agg (PNG output) path.simplify: (true|false) => somewhat
visible borders between polygons (esp. adjacent to contrasting colors)
http://zaneselvans.org/dropbox/contourf_simplifytrue.png
http://zaneselvans.org/dropbox/contourf_simplifyfalse.png

backend: pdf (PDF output) path.simplify: true => reliably visible
irregularities (but probably this is somewhat expected with SVG
output)
http://zaneselvans.org/dropbox/contourf_simplifytrue.pdf

backend: pdf (PDF output) path.simplify: false => infinitessimally
thin lines of background color visible between contour filled regions.
http://zaneselvans.org/dropbox/contourf_simplifyfalse.pdf

And here's what I used to generate them:

def broken_contourf():
    """
    Simple demonstration that filled contour plots are broken.

    Looking at the PDF and PNG output using the macosx backend, everything is
    fine. It's only the GUI output which is for some reason outlining the
    polygons in the filled contours.

    """
    from numpy.random import uniform, seed
    from matplotlib.mlab import griddata
    import matplotlib.pyplot as plt
    import numpy as np
    # make up data.
    #npts = int(raw_input('enter # of random points to plot:'))
    seed(-1)
    npts = 200
    x = uniform(-2,2,npts)
    y = uniform(-2,2,npts)
    z = x*np.exp(-x**2-y**2)
    # define grid.
    xi = np.linspace(-2.1,2.1,100)
    yi = np.linspace(-2.1,2.1,100)
    # grid the data.
    zi = griddata(x,y,z,xi,yi)
    # contour the gridded data, plotting dots at the nonuniform data points.
    CS = plt.contourf(xi,yi,zi,15,cmap=plt.cm.jet)
    # plot data points.
    plt.title('griddata() and contourf() test')
    plt.savefig('broken_contourf.pdf')
    plt.savefig('broken_contourf.png')

···

On Fri, Jun 12, 2009 at 3:02 PM, Zane Selvans<zane@...1923...> wrote:

I switched back to using the macosx backend, and it turns out that the
thin black lines surrounding the polygons (including crossing the
filled contour regions from one closed contour to another) only get
displayed in the GUI. PDF and PNG output look fine.

Zane

On Fri, Jun 12, 2009 at 2:27 PM, Zane Selvans<zane@...1923...> wrote:

If I set path.simplify: False, the shape of the gaps between the
filled polygons does change. Instead of being irregular, it becomes
an infinitessimally thin gap of uniform width, allowing the (in this
case white) background to show through.

In both of these cases (path.simplify: True|False), the PNG version of
the same figures also show representations of these gaps which are
identical to those which appear in the PDF (though obviously
pixelated), so I don't think it's something that's wrong in the vector
graphics code per se.

Zane

On Fri, Jun 12, 2009 at 11:46 AM, Michael Droettboom<mdroe@...86...> wrote:

Shot in the dark here, but what if you set the rcParam "path.simplify" to
False? There have been recent changes to that code.

Also, since the Agg backend doesn't have an associated GUI, you need to use
the savefig() command and provide a filename, rather than using show().

Cheers,
Mike

Zane Selvans wrote:

Um, yeah. So my response got bounced because of the attachment. Take 2:

For some reason my script bombed when I switched to the Agg backend,
trying to display to the screen (it said Figure has no method show())

So I output the plot as both a PDF and a PNG (still having backend:
agg in my rcfile) and in both of those cases, irregular gaps are
visible between the polygons making up the filled contours. This
wasn't the case with my previously installed setup. It looks as if
for some reason the vertices of the filled polygons are being
calculated differently from different sides of the same contour,
leading to overlap in some places, and gaps in others. You can download
the PDF version (in which the exact geometry is much clearer).
from:

http://zaneselvans.org/dropbox/LinDensity_Grid.pdf

Zane

On Fri, Jun 12, 2009 at 5:51 AM, Michael Droettboom<mdroe@...86...> >>>> wrote:

So you see this behavior if you switch to the Agg backend? That's the
backend used to generate the images in the gallery. If there's a
difference
there, that would seem to suggest some tweaking of the macosx backend
(which
is still relatively new) is in order.

Mike

Zane Selvans wrote:

I just installed the latest SciPy Superpack in order to get access to
the scipy.spatial.KDTree class, and discovered that for some reason
now when I use contourf() lines get drawn at the boundaries between
the filled contours. Additionally, there is always a single vertical
line crossing from each contour boundary to the next. I'm guessing
that these are the edges of the filled polygons which are getting
drawn. This behavior doesn't seem to be consistent with the
contourf() documentation and when I run code in griddata_demo.py it
doesn't come out looking like the picture in the documentation/example
gallery...

Is anyone else seeing this behavior? Is there a keyword I can use to
force the edges of the polygons not to get drawn?

This is on Mac OS X 10.5.7, with
scipy.__version__ = 0.8.0.dev5635
matplotlib.__version__ = 0.98.6svn
numpy.__version__=1.4.0.dev6728

As installed by superpack_2009.03.28.sh
from http://macinscience.org/?page_id=6

using:
backend: macosx

Cheers,
Zane

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Michael Droettboom
Science Software Branch
Operations and Engineering Division
Space Telescope Science Institute
Operated by AURA for NASA

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

--
Zane A. Selvans
Amateur Earthling
http://zaneselvans.org
+1 303 815 6866

Thanks for the thorough investigation.

Michiel: can you look into why the macosx backend is drawing the
strokes around the polygons? Probably as simple as inadvertently
ignoring an argument.

I’ll look into the path simplification-related issues.

Cheers,

Mike

···

http://zaneselvans.org/dropbox/contourf_backendmacosx.pnghttp://zaneselvans.org/dropbox/contourf_simplifytrue.pnghttp://zaneselvans.org/dropbox/contourf_simplifyfalse.pnghttp://zaneselvans.org/dropbox/contourf_simplifytrue.pdfhttp://zaneselvans.org/dropbox/contourf_simplifyfalse.pdf<zane@…1923…><zane@…1923…><mdroe@…86…>http://zaneselvans.org/dropbox/LinDensity_Grid.pdf<mdroe@…86…>http://macinscience.org/?page_id=6http://zaneselvans.orghttp://zaneselvans.org