Thread: [Matplotlib-users] Problems with, LaTeX symbols, axes rescaling

Yesterday I compiled Matplotlib from SVN, and installed it to run on  
top of the dependencies that I have installed from the SciPy Superpack  
for OS X from Chris Fonnesbeck.  Nearly everything seems to be working  
fine.  However, there are a couple of glitches, and I can't tell  
whether it's because of something being out of sync between all my  
installed bits, or whether it's just the fact that I'm using the SVN  
code.

The first thing I noticed is that all of the LaTeX symbols I have in  
my plots are now messed up.  I could see this being a font issue...  
does anyone know how/where Matplotlib gets pointed at the LaTeX  
fonts?  And how I would check to see where it's looking, or whether  
it's not finding what it's looking for?  I'm not getting any errors -  
my greek letters just cease to be greek, and various text decorations  
(like \bar{D}) come out as other things altogether (like D, only with  
an Angstrom symbol over it...).

The second thing is that in a figure in which I'm using twinx() to  
create a separate y-axis using the same x-axis, I no longer seem to be  
able to control the limits of the new axes.  They just automatically  
re-scale to fit whatever data is being plotted, regardless of whether  
I use ax.set_ylim().

Third, and this was a problem before I did the re-compile, and it  
remains so, often (but not always) my figures refuse to draw  
completely, until I've re-sized the window in which they're being  
displayed.  This happens even when I do a figure.show() at the end of  
the script.  Some of the lines will draw, but not all of them.  Is  
this a familiar problem?

Thanks for any suggestions...

Zane

--
Zane Selvans
Amateur Earthling
za...@id...
303/815-6866
http://zaneselvans.org
PGP Key: 55E0815F

Zane Selvans wrote:
> Yesterday I compiled Matplotlib from SVN, and installed it to run on top 
[...]
> 
> The second thing is that in a figure in which I'm using twinx() to 
> create a separate y-axis using the same x-axis, I no longer seem to be 
> able to control the limits of the new axes.  They just automatically 
> re-scale to fit whatever data is being plotted, regardless of whether I 
> use ax.set_ylim().

Formerly, autoscaling was not supported by shared axes, which would have 
included twinx.  Now autoscaling is supported.  I suspect this is the 
difference you are seeing.  The solution is to turn off autoscaling with 
the ax.set_autoscale_on method, or to call ax.set_ylim after you have 
plotted your lines.

Eric

On Mon, Nov 24, 2008 at 6:23 PM, Zane Selvans <za...@id...> wrote:

> whether it's just the fact that I'm using the SVN code.
> The first thing I noticed is that all of the LaTeX symbols I have in my
> plots are now messed up.  I could see this being a font issue... does anyone
> know how/where Matplotlib gets pointed at the LaTeX fonts?  And how I would
> check to see where it's looking, or whether it's not finding what it's
> looking for?  I'm not getting any errors - my greek letters just cease to be
> greek, and various text decorations (like \bar{D}) come out as other things
> altogether (like D, only with an Angstrom symbol over it...).

I suggest completely wiping your .matplotlib directory (saving only
your matplotlibrc if you have customized it).  matplotlib caches a
fair amount of font and latex rendering stuff.  I had a similar
problem on an upgrade with mathtext.  We are not sure why it is
happening yet, but it will probably go away if you rm -rf
~/.matplotlib.  Also, wipe your site-packages/matplotlib dir before
installing.  See

  http://matplotlib.sourceforge.net/faq/installing_faq.html#cleanly-rebuild-and-reinstall-everything

To get more verbose information about what mpl is doing behind the
scense, run with


   > python myscript.py --verbose-debug

You can use --verbose-helpful for less output.

I think it might be a good idea for us to start using a versioned dir
structure for the cached data...


JDH

On Nov 24, 2008, at 5:23 PM, John Hunter wrote:

>> The first thing I noticed is that all of the LaTeX symbols I have  
>> in my
>> plots are now messed up.  I could see this being a font issue...  
>> does anyone

> I suggest completely wiping your .matplotlib directory (saving only
> your matplotlibrc if you have customized it).  matplotlib caches a

Thanks, this fixed it!

Though, futzing with the LaTeX settings in the new matplotlibrc it  
appears that something about how/when it chooses when to format as tex  
has changed.  If text.usetex is True, text in my labels and titles  
that is enclosed within $$'s gets typeset as tex, but apparently *all*  
of the xtick labels get set as mathtext - they're in the light/thin  
and slightly more difficult to read LaTeX font, and in my Basemap  
figures, the NSEW modifiers on the text objects indicating lat/lon  
direction are all italicised, as they would be were they enclosed in $ 
$'s.  I don't seem to be able to get the relatively bold, sans-serif  
tick labels, while still preserving the ability to typeset math in my  
labels and titles...

(An aside: how does one go about decorating the Text objects that  
label the ticks?  I frequently have an axis whose units are degrees,  
and I'd rather just be able to append a $^\circ$ on the end of the  
text for the ticks, rather than have to say in the axis label  
[degrees].)

If I set text.markup to 'tex', instead of 'plain' I get a complaint  
about it being invalid, with a pointer to a URL that no longer exists  
to obtain a new matplotlibrc:

Bad key "text.markup" on line 162 in /Users/zane/.matplotlib/ 
matplotlibrc.
You probably need to get an updated matplotlibrc file from http://matplotlib.sf.net/matplotlibrc 
  or from the matplotlib source distribution

Thanks again!
Zane

--
Zane Selvans
Amateur Earthling
za...@id...
303/815-6866
http://zaneselvans.org
PGP Key: 55E0815F

On Mon, Nov 24, 2008 at 8:37 PM, Zane Selvans <za...@id...> wrote:
>
> On Nov 24, 2008, at 5:23 PM, John Hunter wrote:
>
> The first thing I noticed is that all of the LaTeX symbols I have in my
>
> plots are now messed up.  I could see this being a font issue... does anyone
>
> I suggest completely wiping your .matplotlib directory (saving only
>
> your matplotlibrc if you have customized it).  matplotlib caches a
>
> Thanks, this fixed it!
> Though, futzing with the LaTeX settings in the new matplotlibrc it appears
> that something about how/when it chooses when to format as tex has changed.
>  If text.usetex is True, text in my labels and titles that is enclosed
> within $$'s gets typeset as tex, but apparently *all* of the xtick labels
> get set as mathtext - they're in the light/thin and slightly more difficult
> to read LaTeX font, and in my Basemap figures, the NSEW modifiers on the
> text objects indicating lat/lon direction are all italicised, as they would
> be were they enclosed in $$'s.  I don't seem to be able to get the
> relatively bold, sans-serif tick labels, while still preserving the ability
> to typeset math in my labels and titles...

We have two modes to render latex -- one is native TeX layout and uses
latex and dvipng, and is superior if you need all of TeX's
capabilities (eg an eqnarray), but the fonts are suboptimal.  The
other is mathtext, which will cover 98% of most users' math needs,
does not require a latex install because it is pure mpl, and uses
nicer fonts .  The docs are here:

  mathtext: http://matplotlib.sourceforge.net/users/mathtext.html
  usetex: http://matplotlib.sourceforge.net/users/usetex.html

I strongly encourage using mathtext over usetex, unless you need some
of the hairy stuff from TeX.

> (An aside: how does one go about decorating the Text objects that label the
> ticks?  I frequently have an axis whose units are degrees, and I'd rather
> just be able to append a $^\circ$ on the end of the text for the ticks,
> rather than have to say in the axis label [degrees].)

Use a custom formatter.  I'm freestyling here, so check the examples
linked to below

  import matplotlib.ticker as ticker
  formatter = ticker.FormatStrFormatter(r'$%1.2f^\circ$')
  ax.xaxis.set_major_formatter(formatter)

The API docs for the tick formatting and locating are here

  http://matplotlib.sourceforge.net/api/ticker_api.html

And code examples matching "formatter" are here

  http://matplotlib.sourceforge.net/search.html?q=codex+formatter

> If I set text.markup to 'tex', instead of 'plain' I get a complaint about it
> being invalid, with a pointer to a URL that no longer exists to obtain a new
> matplotlibrc:
> Bad key "text.markup" on line 162 in /Users/zane/.matplotlib/matplotlibrc.
> You probably need to get an updated matplotlibrc file
> from http://matplotlib.sf.net/matplotlibrc or from the matplotlib
> source distribution

I've updated the code in svn to point to the new location

  http://matplotlib.sourceforge.net/_static/matplotlibrc

See also http://matplotlib.sourceforge.net/users/customizing.html

JDH

> We have two modes to render latex -- one is native TeX layout and uses
> latex and dvipng, and is superior if you need all of TeX's
> capabilities (eg an eqnarray), but the fonts are suboptimal.  The
> other is mathtext, which will cover 98% of most users' math needs,
> does not require a latex install because it is pure mpl, and uses
> nicer fonts .

Well, if I turn off usetex, then the LaTeX symbols don't get rendered  
correctly.  It's obviously trying to do something other than just  
write out the plain text, but it's not doing the right thing.  I quit  
out of matplotlib, and re-purged all the cached stuff from  
my .matplotlib directory, and it's still failing.  From the docs, it  
sounds like mathtext is just the default way for Text objects to get  
rendered.  It fails the same way irrespective of whether I use a raw  
string, or a double-quoted string:

fit_ax.set_ylabel(r'$\delta_{rms}$ [degrees]')
fit_ax.set_ylabel("$\delta_{rms}$ [degrees]")

both give a capital 'C' with a subscript 'rms'

$\bar{D}$ results in a capital D with a little circle over it.

$20^\circ$ yields 20 raised to the little tiny 'e' power.

Even if I wipe my site-packages/matplotlib* directory and egg-info,  
make clean, and recompile and install matplotlib from the SVN source,  
I still get the same problem.  I don't know if it means anything, but  
I note that the mpl-data/fonts directory in SVN has 282 sub-entries,  
and the mpl-data/fonts directory that's getting installed in my site- 
packages only has 99 sub-entries.  But maybe they aren't supposed to  
be the same?  Dunno.

So for now I can only get my math symbols to render with text.usetext  
turned on.

I'll read up on the text formatters.  Thanks for all your help!

Zane

--
Zane Selvans
Amateur Earthling
za...@id...
303/815-6866
http://zaneselvans.org
PGP Key: 55E0815F

I'm seeing this here, too.  Something must have been messed up recently 
in SVN.  I'm looking into it.

Mike

Zane Selvans wrote:
>> We have two modes to render latex -- one is native TeX layout and uses
>> latex and dvipng, and is superior if you need all of TeX's
>> capabilities (eg an eqnarray), but the fonts are suboptimal.  The
>> other is mathtext, which will cover 98% of most users' math needs,
>> does not require a latex install because it is pure mpl, and uses
>> nicer fonts .
>
> Well, if I turn off usetex, then the LaTeX symbols don't get rendered 
> correctly.  It's obviously trying to do something other than just 
> write out the plain text, but it's not doing the right thing.  I quit 
> out of matplotlib, and re-purged all the cached stuff from my 
> .matplotlib directory, and it's still failing.  From the docs, it 
> sounds like mathtext is just the default way for Text objects to get 
> rendered.  It fails the same way irrespective of whether I use a raw 
> string, or a double-quoted string:
>
> fit_ax.set_ylabel(r'$\delta_{rms}$ [degrees]')
> fit_ax.set_ylabel("$\delta_{rms}$ [degrees]")
>
> both give a capital 'C' with a subscript 'rms'
>
> $\bar{D}$ results in a capital D with a little circle over it.
>
> $20^\circ$ yields 20 raised to the little tiny 'e' power.
> Even if I wipe my site-packages/matplotlib* directory and egg-info, 
> make clean, and recompile and install matplotlib from the SVN source, 
> I still get the same problem.  I don't know if it means anything, but 
> I note that the mpl-data/fonts directory in SVN has 282 sub-entries, 
> and the mpl-data/fonts directory that's getting installed in my 
> site-packages only has 99 sub-entries.  But maybe they aren't supposed 
> to be the same?  Dunno.
> So for now I can only get my math symbols to render with text.usetext 
> turned on.
I'm seeing

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

I introduced this bug trying to fix the Cairo backend on Nov 12. 
(r6400).  It should now be fixed in SVN r6446.  You will need to remove 
~/.matplotlibrc/fontList.cache for the fix to take effect.

Mike

Michael Droettboom wrote:
> I'm seeing this here, too.  Something must have been messed up recently 
> in SVN.  I'm looking into it.
>
> Mike
>
> Zane Selvans wrote:
>   
>>> We have two modes to render latex -- one is native TeX layout and uses
>>> latex and dvipng, and is superior if you need all of TeX's
>>> capabilities (eg an eqnarray), but the fonts are suboptimal.  The
>>> other is mathtext, which will cover 98% of most users' math needs,
>>> does not require a latex install because it is pure mpl, and uses
>>> nicer fonts .
>>>       
>> Well, if I turn off usetex, then the LaTeX symbols don't get rendered 
>> correctly.  It's obviously trying to do something other than just 
>> write out the plain text, but it's not doing the right thing.  I quit 
>> out of matplotlib, and re-purged all the cached stuff from my 
>> .matplotlib directory, and it's still failing.  From the docs, it 
>> sounds like mathtext is just the default way for Text objects to get 
>> rendered.  It fails the same way irrespective of whether I use a raw 
>> string, or a double-quoted string:
>>
>> fit_ax.set_ylabel(r'$\delta_{rms}$ [degrees]')
>> fit_ax.set_ylabel("$\delta_{rms}$ [degrees]")
>>
>> both give a capital 'C' with a subscript 'rms'
>>
>> $\bar{D}$ results in a capital D with a little circle over it.
>>
>> $20^\circ$ yields 20 raised to the little tiny 'e' power.
>> Even if I wipe my site-packages/matplotlib* directory and egg-info, 
>> make clean, and recompile and install matplotlib from the SVN source, 
>> I still get the same problem.  I don't know if it means anything, but 
>> I note that the mpl-data/fonts directory in SVN has 282 sub-entries, 
>> and the mpl-data/fonts directory that's getting installed in my 
>> site-packages only has 99 sub-entries.  But maybe they aren't supposed 
>> to be the same?  Dunno.
>> So for now I can only get my math symbols to render with text.usetext 
>> turned on.
>>     
> I'm seeing
>
>   

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