matplotlib-users Mailing List for matplotlib

>>>>> "Christopher" == Christopher Fonnesbeck <ch...@fo...> writes:

    Christopher> ... I'm not sure why I'd want to use gtk instead of
    Christopher> wx anyhow, given the choice.

One more option.  If all you want is the batch generation of figures
offline with no display, you can use either the PS or GD backends.  PS
only requires Numeric.  gd, however, takes some energy to build.

But it works: I just built all the prereqs on OSX from source and it
worked perfectly.

You need

libjpeg - http://freshmeat.net/redir/libjpeg/5665/url_tgz/jpegsrc.v6b.tar.gz

    ./configure
    sudo make
    sudo cp libjpeg.a /usr/local/lib/
    sudo cp *.h /usr/local/include/

libpng - http://umn.dl.sourceforge.net/sourceforge/libpng/libpng-1.2.5.tar.gz

    cp scripts/makefile.macosx Makefile
    sudo make install


You then need to follow the build instructions for gd, gdmodule and
dependencies at

http://matplotlib.sourceforge.net/backends.html#GD

Here's an example script ....

import matplotlib
matplotlib.use('GD')
from matplotlib.matlab import *

t = arange(0.0, 3.0, 0.01)
for i in range(1,10):
    figure(1)
    s = sin(2*pi*i*t)
    plot(t,s)
    savefig('plot%02d' % i)
    close(1)

I put all this here mainly for archival purposes in case someone wants
to use matplotlib / OSX / gdmodule.  We'll get the wx thing figured
out too ....

JDH

>>>>> "Christopher" == Christopher Fonnesbeck <ch...@fo...> writes:

    Christopher> ... I'm not sure why I'd want to use gtk instead of
    Christopher> wx anyhow, given the choice.

No compelling reason.  The GTK backend was the first and probably the
most stable for that reason.  This is the first release with the wx
backend, so there are likely some bugs lurking that will have to be
found and stamped out.

    Christopher> I'm trying to keep my plotting in a separate module
    Christopher> from everything else, so that I don't have to make
    Christopher> changes all over the place when I change my plotting
    Christopher> code. For example, plot_histogram() can be called
    Christopher> from any other module without knowing how the plot is
    Christopher> implemented. If I have to call show() from each
    Christopher> module that uses the plotting module, it will start
    Christopher> to get messy. Outside of the Plot module, the rest of
    Christopher> my code shouldn't have to know anything about
    Christopher> matplotlib. So, I'd like to be able to gererate and
    Christopher> save a plot, then move on to the next piece of
    Christopher> code. I'm trying to use your package in place of
    Christopher> scipy, which is what I have been using to generate
    Christopher> plots up until now; hopefully I can make it work.

Would you like this to run 'offline' so that no windows pop up as you
create the figures, or is it enough simply that you can create them
one at a time in batch mode with the plot functionality encapsulated?

I just installed wxpython and matplotlib on OS X, so I'll have a
chance to delve into this issue a bit....

John Hunter

On Dec 15, 2003, at 3:29 PM, John Hunter wrote:
>
>
>     Christopher> I'm trying to use matplotlib on OSX (so with WX
>     Christopher> rather than gtk) in conjunction with my own code that
>
> Not necessarily.  pygtk, gtk-2, etc... have been packaged for fink.
> I'm going to try and install them on my powerbook when I get a few
> minutes.

I like to avoid fink unless I absolutely have to use it -- it ends up 
installing a whole bunch of packages that I already have elsewhere on 
my machine, thereby creating the need to maintain multiple 
installations of the same package.

... I'm not sure why I'd want to use gtk instead of wx anyhow, given 
the choice.

>
> show must be the last line of your script.  Is it possible to do all
> the figures in the loop and then call show?  If your number of figures
> is very large, you could run into memory problems this way.  There is
> a work-around, but if you can do it all in memory that is the easiest
> solution
>
> for i in range(1,10):
>     figure(i)
>     # do plot i
>     savefig('plot%d'%i)
> show()
>
> This is an area that I am actively working on (in fact I was working
> on it when you emailed!) so if this isn't a viable solution for you
> let me know.  I'm implementing some features to make it easier to use
> matplotlib in with xvfb (virtual x) so you can produce plots in wx or
> gtk without launching the GUI windows.
>
>

I'm trying to keep my plotting in a separate module from everything 
else, so that I don't have to make changes all over the place when I 
change my plotting code. For example, plot_histogram() can be  called 
from any other module without knowing how the plot is implemented. If I 
have to call show() from each module that uses the plotting module, it 
will start to get messy. Outside of the Plot module, the rest of my 
code shouldn't have to know anything about matplotlib. So, I'd like to 
be able to gererate and save a plot, then move on to the next piece of 
code. I'm trying to use your package in place of scipy, which is what I 
have been using to generate plots up until now; hopefully I can make it 
work.


--
Christopher J. Fonnesbeck ( c h r i s @ f o n n e s b e c k . o r g )
Georgia Cooperative Fish & Wildlife Research Unit, University of Georgia

>>>>> "Christopher" == Christopher Fonnesbeck <ch...@fo...> writes:

    Christopher> code for some reason, but instead hangs. When I try
    Christopher> and close the plot manually, it terminates the python
    Christopher> code. How can I get matplotlib to produce a new plot
    Christopher> each time it is called, moving on after each plot is
    Christopher> produced, displayed and saved to file. Here is a
    Christopher> sample function from my code:

For the GTK backend, there is another solution.  The call to 
ShowOn().set(1) tells matplotlib to execute the commands as they are
issued, which was originally designed for people using matplotlib
from the python shell but also applies to your case, where you want to
generate a bunch of figures one at a time


import matplotlib
matplotlib.use('GTK')
from matplotlib.matlab import *
from matplotlib.backends.backend_gtk import ShowOn
ShowOn().set(1)  

t = arange(0.0, 3.0, 0.01)
for i in range(1,10):
    figure(1)
    s = sin(2*pi*i*t)
    plot(t,s)
    savefig('plot%02d' % i)
    close(1)


*Note there is no call to show at the end*

The same trick *does not* currently work with the WX backend, though
perhaps Jeremy can take a look at it and determine why not.  Somehow
the call the ShowOn().set(1) seems to screw it up.

We'll look into it...

JDH

>>>>> "Christopher" == Christopher Fonnesbeck <ch...@fo...> writes:

    Christopher> First off, I want to congratulate the authors of
    Christopher> matplotlib for making a great package. This is the
    Christopher> python plotting library I've been waiting for.

Thanks!

    Christopher> I'm trying to use matplotlib on OSX (so with WX
    Christopher> rather than gtk) in conjunction with my own code that

Not necessarily.  pygtk, gtk-2, etc... have been packaged for fink.
I'm going to try and install them on my powerbook when I get a few
minutes.

    Christopher> produces Markov chain Monte Carlo simulations. At the
    Christopher> end of the simulation, I try to plot traces and
    Christopher> histograms of all of the sampled values. However,
    Christopher> matplotlib gets stuck on the first plot. After
    Christopher> calling show(), it does not move to the next line of
    Christopher> code for some reason, but instead hangs. When I try
    Christopher> and close the plot manually, it terminates the python
    Christopher> code. How can I get matplotlib to produce a new plot
    Christopher> each time it is called, moving on after each plot is
    Christopher> produced, displayed and saved to file. Here is a
    Christopher> sample function from my code:

show must be the last line of your script.  Is it possible to do all
the figures in the loop and then call show?  If your number of figures
is very large, you could run into memory problems this way.  There is
a work-around, but if you can do it all in memory that is the easiest
solution

for i in range(1,10):
    figure(i)
    # do plot i
    savefig('plot%d'%i)
show()

This is an area that I am actively working on (in fact I was working
on it when you emailed!) so if this isn't a viable solution for you
let me know.  I'm implementing some features to make it easier to use
matplotlib in with xvfb (virtual x) so you can produce plots in wx or
gtk without launching the GUI windows.

JDH

First off, I want to congratulate the authors of matplotlib for making  
a great package. This is the python plotting library I've been waiting  
for.

I'm trying to use matplotlib on OSX (so with WX rather than gtk) in  
conjunction with my own code that produces Markov chain Monte Carlo  
simulations. At the end of the simulation, I try to plot traces and  
histograms of all of the sampled values. However, matplotlib gets stuck  
on the first plot. After calling show(), it does not move to the next  
line of code for some reason, but instead hangs. When I try and close  
the plot manually, it terminates the python code. How can I  get  
matplotlib to produce a new plot each time it is called, moving on  
after each plot is produced, displayed and saved to file. Here is a  
sample function from my code:


def  
plot_histogram(data,name,nbins=None,xlab='Value',ylab='Frequency',suffix 
=''):
	'Internal histogram specification for handling nested arrays'
	
	'If there is only one data array, go ahead and plot it ... '
	if len(shape(data))==1:
		print 'Generating histogram of',name
		try:
			figure(1)
			nbins = min(nbins or 10,len(unique(data)))
			'Generate histogram'
			hist(data,nbins)
			'Plot options'
			xlabel(name)
			ylabel("Frequency")
			show()
			'Save to file'
			savefig("%s%s.png" % (name,suffix))
			close(1)
		except OverflowError:
			print '... cannot generate histogram'
	else:
		'... otherwise, plot recursively'
		tdata = swapaxes(data,0,1)
		for i in range(len(tdata)):

Thanks for any advice,
C.

--
Christopher J. Fonnesbeck ( c h r i s @ f o n n e s b e c k . o r g )
Georgia Cooperative Fish & Wildlife Research Unit, University of Georgia

matplotlib is a pure python 2D plotting library with a matlab syntax
which produces publication quality figures using in a variety of
hardcopy formats (PNG, JPG, TIFF, PS) and interactive GUI environments
(WX, GTK) across platforms.  matplotlib can be used in python scripts,
interactively from the python shell (ala matlab or mathematica), in
web application servers generating dynamic charts, or embedded in GTK
or WX applications.

  http://matplotlib.sourceforge.net

What's new

WX python backend

  Jeremy O'Donoghue has done an amazing job implementing the backend
  for wxpython, with all the features such as interactive navigation
  that are supported in the GTk backend.  matplotlib should not be
  available with a GUI anywhere wxpython, Numeric and python work,
  including linux and friends, win32, and Mac OS X.  See
  http://matplotlib.sourceforge.net/matplotlib.backends.backend_wx.html
  for a summary of known issues.

Pseudo color plots

  The pcolor command generates pseudo color plots.  See
  http://matplotlib.sourceforge.net/screenshots.html#pcolor_demo and
  http://matplotlib.sourceforge.net/screenshots.html#mri_with_eeg for
  screenshots and example code

New time-series plotting functions
  
  * psd - plots the power spectral density of a time series
  * csd - plots the cross spectral density of two time series
  * cohere - plots the coherence

  See the examples dir in the src distro: psd_demo.py and csd_demo.py

Substantially improved layout

  The transform architecture was refactored, allowing much more
  precise layout.  Lines, patches, text, etc... can now be placed and
  scaled in arbitrary units, relative axes units, or physical size.
  Applicaiton programmers who want to create lines, patches and text
  directly using the API should read the transform module docs for
  more info.  See the text help for an example of how to specify text
  locations in axes coords (0,0 is lower left and 1,1 is upper right)

Expanded legend capabilities

  The legend class is improved, with a more sophisticated layout
  engine and the ability to accept lines and rectangle patches as an
  optional first argument to specify which lines/patches make up the
  legend.  There are also additional legend placement locations, like
  'upper center'.  See
  http://matplotlib.sourceforge.net/screenshots.html#legend_demo

Expanded errorbar capabilities

  Gary Ruben contributed some code to support x and y errorbars,
  either symmetrix or asymettric, in one enhanced function 'errorbar'.
  See the errorbar_demo.py for examples of all the wild and wonderful
  errorbar styles.  Bar charts can now also display errorbars; see
  http://matplotlib.sourceforge.net/screenshots.html#barchart_demo

Figure size and DPI controllable from matlab interface

  The figure command now takes optional args figsize and dpi to set
  the figure size and DPI in the matlab interface.  This change
  involved some changes in the way default sizes and resolutions were
  handled among the various backends, with the effect that figures
  generated by existing scripts may appear different, eg, in the
  relative size of text to the figure elements.  The advantage is that
  the current implementation does a better job of computing true sizes
  with increased fidelity between backends.  Apologies for any
  inconveniences!

API changes

  There have been some minor changes to the API for those using
  matplotlib embedded in GTK applications.  

  * If you instantiate an Axes of Subplot intace, the first arg to
    __init__ must be the figure that contains it.

  * If you instantiate any artists, eg, Line2D, Rectangle, or
    AxisTextGTk, you must initialize them with their dpi, bbox, and
    transforms.  See the help for the transforms modules and the
    examples logo.py and mri_with_eeg.py, where the objects are
    explicity created using the new API.

  * The only change to the matlab interface is in the signature of the
    errorbar func, which breaks matlab compatibility for the enhanced
    ability to do x and y errorbars.

  * See matplotlib.axes.py for a complete list of API changes


John Hunter

>>>>> "K" == K KISHIMOTO <ko...@us...> writes:

    K> I am sorry that my last explanation was insufficient.  The
    K> point is that I think it is inefficient to execute the
    K> following two lines in matplotlib.mlab.hist when I know the
    K> result of "n" already.

Good point ...

    K> Another point is hist() does not support error bar plot (but
    K> bar() called from hist() does support.)

Another good point....

    K> So, I propose again def hist(x, bins=10, noplot=0, normed=0,
    K> weights=None, errors=None , **kwargs): in matplotlib.matplab.
    K> I think if matplotlib supports this by default, it is very
    K> smart and usefull to many users when making histogram plots
    K> with matplotlib than letting hist() be mainly for the purpose
    K> of "calculating" histograms.

OK, I'll give it some thought.  Sounds reasonable enough.  It doesn't
break backwards compatibility, adds useful features, and is more
consistent with bar.

JDH

Thank you for your reply.

> def func1(x=None):
>     if x is None: x = []
>     #blah blah

The indication is right.
I should have done so. Thank you.

> This seems like the kind of thing that would best be done in your user
> library.  Eg, if you make a module mymatplotlib.py, you can defined
> your own hist.  In that file, just import matplotlib.matlab and call
> matplotlib.matlab.hist within it.  In fact, matplotlib.matlab.hist
> calls matplotlib.mlab.hist.

I am sorry that my last explanation was insufficient.
The point is that I think it is inefficient to execute the following 
two lines in matplotlib.mlab.hist when I know the result of "n" already.
 >> n = searchsorted(sort(y), bins)
 >> n = diff(concatenate([n, [len(y)]]))

For the moment, I must insert extra code
 >> y = [1.0] * 1000
(and hist(y, [0.0, 2.0]) calculates n = [1000], then plot the data.)
This is very expensive.

Another point is hist() does not support error bar plot (but bar() 
called from hist() does support.)

So, I propose again
def hist(x, bins=10, noplot=0, normed=0, weights=None, errors=None , 
**kwargs):
in matplotlib.matplab.
I think if matplotlib supports this by default, it is very smart and 
usefull to many users when making histogram plots with matplotlib than 
letting hist() be mainly for the purpose of "calculating" histograms.

Example rough code.
------
def hist(x, bins=10, noplot=0, normed=0, weights=None, xerr=None, 
yerr=None, **kwargs):
    if noplot:
       if weights == None:
          return mlab.hist(x, bins, normed)
       else:
          return (weights, bins)
    else:
       try:
          if weights == None:
             ret = gca().hist(x, bins, normed)
          else:
             ret = gca().bar(bins, weights, xerr=xerr, yerr=yerr, 
**kwargs)
       except ValueError, msg:
          msg = raise_msg_to_str(msg)
          error_msg(msg)
          raise RuntimeError, msg
    draw_if_interactive()
    return ret