How can I open the interactive matplotlib window in IPython notebook?

According to the documentation, you should be able to switch back and forth like this:

In [2]: %matplotlib inline 
In [3]: plot(...)

In [4]: %matplotlib qt  # wx, gtk, osx, tk, empty uses default
In [5]: plot(...) 

and that will pop up a regular plot window (a restart on the notebook may be necessary).

If all you want to do is to switch from inline plots to interactive and back (so that you can pan/zoom), it is better to use %matplotlib magic.

#interactive plotting in separate window
%matplotlib qt 

and back to html

#normal charts inside notebooks
%matplotlib inline 

%pylab magic imports a bunch of other things and may even result in a conflict. It does "from pylab import *".

You also can use new notebook backend (added in matplotlib 1.4):

#interactive charts inside notebooks, matplotlib 1.4+
%matplotlib notebook 

If you want to have more interactivity in your charts, you can look at mpld3 and bokeh. mpld3 is great, if you don't have ton's of data points (e.g. <5k+) and you want to use normal matplotlib syntax, but more interactivity, compared to %matplotlib notebook . Bokeh can handle lots of data, but you need to learn it's syntax as it is a separate library.

Also you can check out pivottablejs (pip install pivottablejs)

from pivottablejs import pivot_ui
pivot_ui(df)

However cool interactive data exploration is, it can totally mess with reproducibility. It has happened to me, so I try to use it only at the very early stage and switch to pure inline matplotlib/seaborn, once I got the feel for the data.

Starting with matplotlib 1.4.0 there is now an an interactive backend for use in the notebook

%matplotlib notebook

There are a few version of IPython which do not have that alias registered, the fall back is:

%matplotlib nbagg

If that does not work update you IPython.

To play with this, goto tmpnb.org

and paste

%matplotlib notebook

import pandas as pd
import numpy as np
import matplotlib

from matplotlib import pyplot as plt
import seaborn as sns

ts = pd.Series(np.random.randn(1000), index=pd.date_range('1/1/2000', periods=1000))
ts = ts.cumsum()

df = pd.DataFrame(np.random.randn(1000, 4), index=ts.index,
                  columns=['A', 'B', 'C', 'D'])
df = df.cumsum()
df.plot(); plt.legend(loc='best')    

into a code cell (or just modify the existing python demo notebook)

A better solution for your problem might be the Charts library. It enables you to use the excellent Highcharts javascript library to make beautiful and interactive plots. Highcharts uses the HTML svg tag so all your charts are actually vector images.

Some features:

• Vector plots which you can download in .png, .jpg and .svg formats so you will never run into resolution problems
• Interactive charts (zoom, slide, hover over points, ...)
• Usable in an IPython notebook
• Explore hundreds of data structures at the same time using the asynchronous plotting capabilities.

Disclaimer: I'm the developer of the library