I would like to use inline plotting in the ipython notebook, i.e.
%matplotlib inline
import matplotlib.pyplot as plt
x = np.arange(100)
plot(x, x**2)
should show an image.
Yet I only get the following message:
lib/python2.7/site-packages/IPython/core/formatters.py:239: FormatterWarning: Exception in image/png formatter: Could not create write struct
FormatterWarning,
What could be the reason for this?
matplotlib==1.3.1 and ipython==2.1.0
Does it say anything on the terminal (i.e. the server)?
My guess is that this is most probably due to some libpng incompatibility issues. If you are running this on OS X, the following discussion may help:
libpng version incompatibility in fresh installation of IPython
Even if you aren't running OS X, similar situations may occur, if you have several copies of libpng floating around.
Related
I recently started using Pycharm instead of Jupyter on my laptop to do some light scientific work(mainly because I like how it auto-completes the code).
this code works fine I Jupyter
import sympy as sp
import matplotlib.pyplot as plt
x = sp.symbols("x")
sp.Integral(x, x)
but when I run it in Pycharm, nothing happens and I don't get any output and I don't know why.
I just upgraded my Jupyter version to 4.2.1 and it seems as though inline figures have gotten a lot larger for the same figsize.
Am I imagining this?
Can I change that without changing the figsize?
You can use top-level matplotlib settings like this:
import matplotlib as mpl
mpl.rcParams['figure.figsize'] = (10,10)
This will change default figure size to 10x10. More on that in the documentation: http://matplotlib.org/users/customizing.html
The scaling of the figure for figures with %matplotlib inline depends on the figure.dpi parameter. So if your figures are to small and you do not want to increase the figsize then you can just set
import matplotlib.pyplot as plt
plt.rcParams['figure.dpi'] = 150 # default for me was 75
I also had the impression that the scaling changed at some point, but I guess this can also appear if you change your screen/resolution.
I am a heavy user of jupyter notebook and, lately, I am running it using pypy instead of python to get extra speed. It works perfectly but I am missing matplotlib so much. Is there any decent 2D plotting library compatible with pypy and jupyter notebook? I don't need fancy stuff, scatter, line and bar plots would be more than enough.
Bokeh is working fairly good with pypy. The only problem I have encountered is linked to the use of numpy.datetime64 that is not yet supported by pypy. Fortunately it is enough to monkey-patch bokeh/core/properties.py and bokeh/util/serialization.py to pass in case of datetime64 reference.
I did it in this way:
bokeh/core/properties.py
...
try:
import numpy as np
datetime_types += (np.datetime64,)
except:
pass
...
and
bokeh/util/serialization.py
...
# Check for astype failures (putative Numpy < 1.7)
try:
dt2001 = np.datetime64('2001')
legacy_datetime64 = (dt2001.astype('int64') ==
dt2001.astype('datetime64[ms]').astype('int64'))
except:
legacy_datetime64 = False
pass
...
And managed to get nice looking plots in jupyter using pypy.
I'd like to learn how to configure the defaults for matplotlib using the inline backend in jupyter notebook. Specifically, I'd like to set default 'figure.figsize’ to [7.5, 5.0] instead of the default [6.0, 4.0]. I’m using jupyter notebook 1.1 on a Mac with matplotlib 1.4.3.
In the notebook, using the macosx backend, my matplotlibrc file is shown to be in the standard location, and figsize is set as specified in matplotlibrc:
In [1]: %matplotlib
Using matplotlib backend: MacOSX
In [2]: mpl.matplotlib_fname()
Out[2]: u'/Users/scott/.matplotlib/matplotlibrc'
In [3]: matplotlib.rcParams['figure.figsize']
Out[3]:[7.5, 5.0]
However, when I use the inline backend, figsize is set differently:
In [1]: %matplotlib inline
In [2]: mpl.matplotlib_fname()
Out[2]: u'/Users/scott/.matplotlib/matplotlibrc'
In [3]: matplotlib.rcParams['figure.figsize']
Out[3]:[6.0, 4.0]
In my notebook config file, ~/.jupyter/jupyter_notebook_config.py, I also added the line
c.InlineBackend.rc = {'figure.figsize': (7.5, 5.0) }
but this had no effect either. For now I’m stuck adding this line in every notebook:
matplotlib.rcParams['figure.figsize']=[7.5, 5.0]
Is there any way to set the default for the inline backend?
The Jupyter/IPython split is confusing. Jupyter is the front end to kernels, of which IPython is the defacto Python kernel. You are trying to change something related to matplotlib and this only makes sense within the scope of the IPython kernel. Making a change to matplotlib in ~/.jupyter/jupyter_notebook_config.py would apply to all kernels which may not make sense (in the case of running a Ruby/R/Bash/etc. kernel which doesn't use matplotlib). Therefore, your c.InlineBackend.rc setting needs to go in the settings for the IPython kernel.
Edit the file ~/.ipython/profile_default/ipython_kernel_config.py and add to the bottom: c.InlineBackend.rc = { }.
Since c.InlineBackend.rc specifies matplotlib config overrides, the blank dict tells the IPython kernel not to override any of your .matplotlibrc settings.
If the file doesn't exist, run ipython profile create to create it.
Using Jupyter on windows at least, I was able to do it using something very much like venkat's answer, i.e.:
%matplotlib inline
import matplotlib
matplotlib.rcParams['figure.figsize'] = (8, 8)
I did this to square the circle, which had been rather eliptical up to that point. See, squaring the circle is not that hard. :)
Note that the path of ipython_kernel_config.py differs if you run ipython from a virtual environment. In that case, dig in the path where the environment is stored.
Use figsize(width,height) in the top cell and it changes width of following plots
For jupyter 5.x and above with IPython kernels, you can just override particular keys and leave the rest by putting things like this, with your desired figsize in your ~/.ipython/profile_default/ipython_kernel_config.py:
c = get_config()
c.InlineBackend.rc.update({"figure.figsize": (12, 10)})
I have the following code in IPython running IPython QT Console on Linux.
%pylab inline
Welcome to pylab, a matplotlib-based Python environment [backend: module://IPython.zmq.pylab.backend_inline].
For more information, type 'help(pylab)'.
fig = figure()
ax = fig.add_axes()
ax = fig.add_axes([0,500, 0, 5000])
ax.plot([1,2,3,44], [4,4,55,55])
Out[5]: [<matplotlib.lines.Line2D at 0x3d8e7d0>]
fig
Out[6]: <matplotlib.figure.Figure at 0x3d25fd0>
fig.show()
/usr/lib/pymodules/python2.7/matplotlib/figure.py:362: UserWarning: matplotlib is currently using a non-GUI backend, so cannot show the figure
"matplotlib is currently using a non-GUI backend, "
I've been struggling to make this work for some time, I've tried changing the backend manually with matplotlib.use() to Qt4Agg, GTK etc with no luck. This also happens in IPython notebook even when I call display().
Any ideas how to get the inline plotting working?
Marked Jakob's answer as the answer, but both are true actually. I had to replace the matploblibrc file with a new copy, started IPython QT Console with --pylab=None then manually entered %pylab inline in the console. Somehow this fixed the problem.
The axis object is defined incorrectly, this prevents matplotlib from rendering.
Remove the first ax = fig.add_axes(), and replace the second line with
ax = fig.add_axes([0, 0, 1, 1]).
The add_axes method requests the size of the axis in relative coordinates, in the form left, bottom, width, height with values between 0 and 1, see e.g. matplotlib tutorial.
You may also try fig.add_subplot(111) instead of fig.add_axes() or fig,ax = subplots() to create your figure and axis objects. The latter assumes that you have populated the interactive namespace matplotlib (%pylab) call in IPython.
It looks like your matplotlib build was compiled without a gui backend.
This is done when either a) it's explicitly specified (handy for webservers), or b) the required libraries for at least one gui backend aren't present (e.g. no Tk, Gtk, Qt, etc).
How did you install matplotlib?
If you compiled it from source, make sure that you have the development libraries for at least Tk installed and that your python install was compiled with Tk support (it is by default). If you installed it from your distro's repositories, whoever built the package built it without gui support, and you'll need to install it from another source.