matplotlib-users Mailing List for matplotlib (Page 3)

John Hunter wrote:

>  > I can confirm that if the font files are installed in the default
>  > location under the system python tree, it works. Setting AFMPATH also
>  > works in linux.
> 
> OK, so the font files exist and are readable.  In order to figure out
> where the problem is, I'll need you to add a couple of lines to
> matplotlib/backends/backend_ps.py and tell me what the output is.  In
> the function _get_afm_filenames on line 413, add these print lines
> before the return statement and let me know what they report
> 
>     print paths
>     print fnames
>     return fnames
> 
> John Hunter

H:\00work\00cvs\matplotlib\matplotlib\examples>python subplot_demo.py -dPS
['E:\\Py23\\share\\matplotlib', 'E', '\\Py23e\\share\\matplotlib']
[]
Error: Could not find any AFM files; please set AFMPATH to point to some 
readable adobe font metric files

Regards,
ST
--

 > I can confirm that if the font files are installed in the default
 > location under the system python tree, it works. Setting AFMPATH also
 > works in linux.

OK, so the font files exist and are readable.  In order to figure out
where the problem is, I'll need you to add a couple of lines to
matplotlib/backends/backend_ps.py and tell me what the output is.  In
the function _get_afm_filenames on line 413, add these print lines
before the return statement and let me know what they report

    print paths
    print fnames
    return fnames

John Hunter

John Hunter wrote:
 >
>     LUK> Another question: I've set the environmental variable AFMPATH
>     LUK> in the registry via the control panel but matplotlib
>     LUK> complains that the .afm files are not there.
> 
> I cannot replicate this problem on my system.  Are the *.afm files in
> E:\Py23e\share\matplotlib?  Are you sure you are running the right
> version of python, ie, do you still get the error with
> 
>   > E:\Py23e\python simple_plot.ps -dPS
> 

[snipped]

> 
> And let me know what it reports.  If you are using the fonts that ship
> with matplotlib, you shouldn't even need to set the AFMPATH because it
> automatically looks in the share/matplotlib subdir of your python
> install dir.  

I can confirm that if the font files are installed in the default 
location under the system python tree, it works. Setting AFMPATH also 
works in linux.

> 
> You may also want to verify that matplotlib is finding your
> environment variable by doing
> 
> 
>>python
>>
>>>>import os
>>>>print os.environ.has_key('AFMPATH')
> 

 >>> import os
 >>> os.environ["AFMPATH"]
'E:\\Py23e\\share\\matplotlib'

> 
> If the *.afm fonts are in the share/matplotlib dir, and you are using
> the right python, and the paths look correct from the diagnostic
> information above, the only thing I can think of is that these files
> are not readable by you (admin versus user thing?). To test this try
> 
> 
>>python
>>
>>>>fh = file('E:\Py23e\\share\\matplotlib\\phvl8a.afm', 'r')
>>>>s = fh.read()
>>>>len(s)
> 

 >>> fh = file('E:\Py23e\\share\\matplotlib\\afm\\phvl8a.afm', 'r')
 >>> len(fh.read())
15627

> 
> Let me know....
> JDH
> 

Regards,
ST
--

>>>>> "Flavio" == Flavio C Coelho <fcc...@fi...> writes:

    Flavio> Hi John et al., I know that the development of a wx
    Flavio> backend is under way.

Well under way in fact.  Jeremy O'Donoghue sent me a version yesterday
which I checked into CVS.  It is still under active development, with
several known bugs, but he has tested it across platforms with most of
the example scripts.  I'll post in the text of his email below which
has more details.

    Flavio> But meanwhile, can any of the non GTK backends (PS and GD)
    Flavio> be used from within a wx app?

    Flavio> I suspect that the answer to this question is no because
    Flavio> the mere import of matplotlib.matlab from any wxpython
    Flavio> shell (like pycrust) will crash it.  But I don't want to
    Flavio> assume that before I am certain of it.

matplotlib should not crash pycrust (it doesn't on my system).  This
may be related to an improper GTK install on your system.  In any
case, you need to import the backend you plan to use before importing
matplotlib.matlab.  I just did the following from pycrust w/o problems

import matplotlib
matplotlib.use('PS')
import os

def psshow():
    savefig('tmp.ps')
    os.system('ggv tmp.ps')
    close()

from matplotlib.matlab import *

plot([1,2,3])
psshow()

t = arange(0.0, 1.0, 0.01)
plot(t, sin(2*pi*t))
psshow()

I also did the the same with GD module.  Just replace 'PS' with GD and
execute your favorite image viewer command.

For either of these, though, You'll first need to comment out line 43
in _matlab_helpers.py, which reads 'figwin.window.destroy()'.  I only
discovered this bug when trying to answer your question.

Now, on to WX.  It is currently only working from pycrust and is under
heavy development, but you are welcome to try it from CVS.  Let me
know if you need a win32 installer and I'll email one.

Here is Jeremy's email:

I thought I'd keep you up to date with my progress.

Attached is the set of files for an early version of the wxWindows
back-end. It seems to work with most of the examples on the website, and
has been tested with wxPython 2.3.3 (on wxWindows 2.2.9) on Windows 2000
using Active State Python 2.2, and (less thoroughly) using wxWindows 2.4
under WxGTK with Python 2.3 on Debian GNU/Linux.

The implementation is very similar to your GTK back-end (probably too
similar, but I like re-using code which works :) This at least means that
it should be pretty easy for you to follow what I've done.

The code is probably only ready for very early users - if anyone is
screaming for wxWindows, you may wish to put it into CVS, otherwise it may
be better to wait for a while - say until I have the interactive functions
working. I'll leave it to your judgement.

Best regards

Jeremy

Known bugs:
- Coloured text labels to not work correctly. I do
  not yet understand why not, and will need to look
  carefully at how wxWindows treats text, as I think
  the code I have is OK!
- I currently do not perform any clipping, as this
  clips text labels.
- Under Windows 2000, the Figure window is larger
  than the figure (seems OK on Linux, however)
- Interactive sessions currently only work inside
  a Python console which already has a wxWindows
  event loop (I recommend PyCrust, which is
  distributed with wxPython)

Not implemented:
- Printing
- Interactive operation using the toolbar (probably
  not a large job to do, as I suspect that I can simply
  re-use the functions from backend_gtk).

Ugly hacks:
- Dotted lines - I have a nasty hack to make this work
  sufficiently for demonstration purposes. The problem
  is that you currently implement dotted lines in a rather
  GTK-specific way. I'd prefer, given that there are only
  a few 'dotted' styles, for these to have symbolic names
  (e.g. in a dictionary), and to do all of the decisions
  on how to draw the 'dotted' style in the back-end.
- As wxWindows has no 'stock' icons, I have converted GTK
  icons similar to those used in backend_gtk into XPM
  format and put them into an 'images' directory under
  backends. I'm open to your preference as to whether it
  would be better to put the images into python code or
  leave them as pixmaps.

Hi John et al.,

I know that the development of a wx backend is under way.

But meanwhile, can any of the non GTK backends (PS and GD) be used from
within a wx app?

I suspect that the answer to this question is no because the mere import
of matplotlib.matlab from any wxpython shell (like pycrust) will crash
it.  But I don't want to assume that before I am certain of it.

do the PS & GD backend support the show()  command? i.e., can they be
easily displayed on the screen or are they geared towards saving and
printing?

thanks

Fl=E1vio Coelho

>>>>> "LUK" == LUK ShunTim <shu...@po...> writes:


    LUK> I followed the link in the matplotlib site and installed gtk
    LUK> 2.2.4.1 from the sf site and the problem goes away. This
    LUK> version has all the images libraries with them. I must
    LUK> bookmark it!

Good news ...

    LUK> Another question: I've set the environmental variable AFMPATH
    LUK> in the registry via the control panel but matplotlib
    LUK> complains that the .afm files are not there.

I cannot replicate this problem on my system.  Are the *.afm files in
E:\Py23e\share\matplotlib?  Are you sure you are running the right
version of python, ie, do you still get the error with

  > E:\Py23e\python simple_plot.ps -dPS

You might want to add a diagnostic debug line to
backends/backend_ps.py to print the paths that it is searching.  After
line 422 in that file, insert a new line that prints the paths, eg,


    def _get_afm_filenames(self):
        paths = [os.path.join(distutils.sysconfig.PREFIX, 'share', 'matplotlib')]
        if os.environ.has_key('AFMPATH'):
            afmpath = os.environ['AFMPATH']
            if afmpath.find(';')>0: #win32 style
                paths.extend(afmpath.split(';'))
            elif afmpath.find(':')>0: # unix style
                paths.extend(afmpath.split(':'))
            else:
                paths.append(afmpath)
***     print paths
        fnames = []

And let me know what it reports.  If you are using the fonts that ship
with matplotlib, you shouldn't even need to set the AFMPATH because it
automatically looks in the share/matplotlib subdir of your python
install dir.  

You may also want to verify that matplotlib is finding your
environment variable by doing

> python
>>> import os
>>> print os.environ.has_key('AFMPATH')

If the *.afm fonts are in the share/matplotlib dir, and you are using
the right python, and the paths look correct from the diagnostic
information above, the only thing I can think of is that these files
are not readable by you (admin versus user thing?). To test this try

> python
>>> fh = file('E:\Py23e\\share\\matplotlib\\phvl8a.afm', 'r')
>>> s = fh.read()
>>> len(s)

Let me know....
JDH

John Hunter wrote:

>>>>>>"LUK" == LUK ShunTim <shu...@po...> writes:
> 
> 
>     LUK> Hello, When I tried out "python simple_plot.py" and tried to
>     LUK> save it as a png file, python crashes. No problem with saving
>     LUK> it to jpeg, though.
> 
> I haven't seen that before.  It works fine on my system (Windows ME)
> and noone else has reported a win32 problem.  The fact that it works
> with jpeg and not png suggests a lib PNG problem.  What version of
> pygtk and GTK-Runtime are you using.  If you are using pygtk-2.0.0 and
> GTK-Runtime-2.2.4.1, then you should be OK.  Have you installed an
> earlier version of these libs, possibly creating a library conflict?
> 
> When you search for libpng, what turns up?
>
> JDH
> 

GTK 2.2.4 (Can't remember where I got it. There's been a few of these 
floating around.) Pygtk 2.0.0 from the link in matplotlib site.

Somehow my GTK runtime don't have the png, jpeg etc images libraries and 
I got them from the gnuwin32 sf site.

I followed the link in the matplotlib site and installed gtk 2.2.4.1 
from the sf site and the problem goes away. This version has all the 
images libraries with them. I must bookmark it!

Another question: I've set the environmental variable AFMPATH in the 
registry via the control panel but matplotlib complains that the .afm 
files are not there.

<quote>
H:\00work\00cvs\matplotlib\matplotlib\examples>python simple_plot.py -dPS
Error: Could not find any AFM files; please set AFMPATH to point to some 
readable adobe font metric files

H:\00work\00cvs\matplotlib\matplotlib\examples>echo %AFMPATH%
E:\Py23e\share\matplotlib
</quote>

This is a W2K system.

Thanks very much,
ST
--

>>>>> "LUK" == LUK ShunTim <shu...@po...> writes:

    LUK> Hello, When I tried out "python simple_plot.py" and tried to
    LUK> save it as a png file, python crashes. No problem with saving
    LUK> it to jpeg, though.

I haven't seen that before.  It works fine on my system (Windows ME)
and noone else has reported a win32 problem.  The fact that it works
with jpeg and not png suggests a lib PNG problem.  What version of
pygtk and GTK-Runtime are you using.  If you are using pygtk-2.0.0 and
GTK-Runtime-2.2.4.1, then you should be OK.  Have you installed an
earlier version of these libs, possibly creating a library conflict?

When you search for libpng, what turns up?

JDH

Hello,

When I tried out "python simple_plot.py" and tried to save it as a png 
file, python crashes. No problem with saving it to jpeg, though.

My system: W2K SP4/Python2.3.2/matplotlib-0.31

Here's some traceback:

<quote>
python.exe caused an Access Violation at location 77f92373 in module 
ntdll.dll Writing to location 00000010.

Registers:
eax=00000000 ebx=00000000 ecx=00000000 edx=00000000 esi=7803a710 
edi=00000000
eip=77f92373 esp=0012e83c ebp=0012e89c iopl=0         nv up ei pl nz na 
pe nc
cs=001b  ss=0023  ds=0023  es=0023  fs=0038  gs=0000 
efl=00000202

Call stack:
77F92373  ntdll.dll:77F92373  RtlpWaitForCriticalSection
77F922F8  ntdll.dll:77F922F8  RtlImageDirectoryEntryToData
03B5C06C  libpng.dll:03B5C06C  png_set_read_fn
01F5F480
00FA5192  libgdk_pixbuf-2.0-0.dll:00FA5192  gdk_pixbuf_new_from_xpm_data
00FA53FF  libgdk_pixbuf-2.0-0.dll:00FA53FF  gdk_pixbuf_savev
00E822DA  _gtk.pyd:00E822DA  init_gtk
1E0591AA  python23.dll:1E0591AA  PyCFunction_Call
</quote>

Regards,
ST
--

A slightly different approach to live plotting.  I have a numerical
program that calls a status function every run through the main loop. It
allows the user to define a status function and pass it that function
(defaulting to just a one-line report on sys.err).

Using GTK idle slowed things down way too much, because GTK was always
idle, even though the number cruncher wasn't. GTK timeout didn't work
either. Using threads was too complicated and error prone, for no gain.
What I wanted was a meter that updated every second or so.

Here is my status function. It shows traces for the best score so far, and
the current score. If enough time has elapsed, it plots (thus scheduling
GTK events), otherwise no. So the gtk.events_pending() call near the end
is usually false.

It rescales the graph too, as required. You can ignore the "interrupted"
stuff -- that allows me to send a "stop" signal to the numerical routine,
saying "Ok, it looks to me like you've converged, stop now."

	-Charles

----------

def my_status_func(*args, **kwargs):

    """    Gather info from a running snob, and plot it.

    It expects the final arg to be a dictionary with
    the following keyword arguments:
          fig   --  the matplot figure for drawing to
          ax    --  the matplot axes in that figure
          t1    --  a list of length 1, with timestamp of last refresh
          iteratns --  sequence of all iteration #s to date (x axis)
          currlens --  sequence of all current lengths to date
          bestlens --  sequence of all best lenghts to date
    All of these kwargs will be modified, at least sometimes.

    """

    #global currlens, bestlens, iteratns, t
    UPDATE_INTERVAL = 1    # how long in s to wait before updating graph

    itr     = args[0] + 1
    max_itr = args[1]
    cl      = int (args[2][0] / N.log(2))
    bl      = int (args[3][0] / N.log(2))
    #cl = int (curr[0] / N.log(2))
    #bl = int (best[0] / N.log(2))


    # Extract variables from kwargs
    t1  = kwargs['t1']
    iteratns = kwargs['iteratns']
    currlens = kwargs['currlens']
    bestlens = kwargs['bestlens']

    iteratns.append (itr)
    currlens.append (cl)
    bestlens.append (bl)

    if itr == 1:
        print "Original length: ", cl

    delta = time.time() - t1[0]

    if (itr > 1) and ( (delta > UPDATE_INTERVAL) or (itr >= max_itr) ):
        # matplot has trouble if we plot just a single data point
        # so we wait for itr > 1

#        if not ShowOn().get():
#            ShowOn().set(True)

        fig = kwargs['fig']
        ax  = kwargs['ax']

        xlo, xhi = ax.get_xlim()
        xlo = max ( 0, xlo)  # Check bc original call is ignored
        xhi = max (50, xhi)  # Likewise
        if itr > xhi:
            xhi *= 2

        ax.set_xlim ([xlo, xhi])
        ax.set_ylim ( [ min(currlens), max(currlens) ] )
        plot(iteratns, currlens, 'b',
             iteratns, bestlens, 'r')


        fig.draw()   # schedule a refresh
        t1[0] = time.time()

    # Display any updates to the figure
    while gtk.events_pending():
        gtk.mainiteration()  # if running in normal gtk thread

    global interrupted
    if interrupted:
        return 1

    return snob.default_status_func(*args)

Pretty!

I just finished adding pseudo-color plots to matplotlib in the CVS
repository.

Here are a couple of screenshots

  http://nitace.bsd.uchicago.edu:8080/files/share/pcolor_demo_large.png

  http://nitace.bsd.uchicago.edu:8080/files/share/mri_with_eeg_large.png

The scripts that generated them are in examples/pcolor_demo.py and
examples/mri_with_eeg.png.

There is currently bug with pcolor in the GTK backend can cause a
vertical banding to appear due to integer rounding in very small
pcolor plots (eg, when you place the plot in a small subplot).  The GD
and PS backends work fine however.

Enjoy!
John Hunter