Python for interactive scientific data visualization

2013-05-15

Some random notes on recent 'pythonic peregrinations' on my Airbook.

Python packages management is really painful. My /Library/Python/2.7/site-packages is just a mess. This is probably due in part to the fact that I switched from easy_install to pip two years ago, but anyway there's a lot of useless stuff in there.

I heard about Bokeh, a new plotting library for Python. Basically, it ought to embed Wilkinson's Grammar of Graphics into the d3js framework. It is maintained by the same guys from Continuum Analytics who develop Blaze and Anaconda. In contrast to Enthought EPD,(a) it is

completely free enterprise-ready Python distribution for large-scale data processing, predictive analytics, and scientific computing.

Anaconda

I followed the instructions to install Anaconda, and it is quite simple actually: You just have to sh Anaconda-*.sh and you should end up with something like:

installing: anaconda-1.4.0-np17py27_0 ...
Python 2.7.3 :: Continuum Analytics, Inc.
creating default environment...
installation finished.

You may wish to edit your .bashrc or prepend the Anaconda install location:

$ export PATH=/Users/chl/anaconda/bin:$PATH

Thank you for installing Anaconda!

WARNING:
    You current have a PYTHONPATH environment variable set. This may cause
    unexpected behavior when running the Python interpreter in Anaconda.
    For best results, please verify that your PYTHONPATH only points to
    directories of packages that are compatible with the Python interpreter
    in Anaconda: /Users/chl/anaconda

Anaconda supports multiple versions of Python to live on your system, and it provides a virtual environment that allow to switch to different version of the same package without much difficulty. Package management is done via conda, a command-line utility that is used to fetch one or more specific versions of a package, display information on the system or environment locations, etc.

% conda search --all numpy
12 matches found:

Packages with available versions and build strings:

numpy                     1.5.1                    py26_0
                          1.5.1                    py27_0
                          1.6.2                    py26_0
                          1.6.2                    py27_0
                          1.7.0b2                  py26_0
                          1.7.0b2                  py27_0
                          1.7.0rc1                 py27_0
                          1.7.0rc1                 py33_0
                          1.7.0rc1                 py26_0
                          1.7.0                    py26_0
                          1.7.0                    py33_0
                          1.7.0                    py27_0

As can be seen, there is a version of numpy 2.7 for Python 3. I haven't explore the full capabilities of this distribution, but at first sight it sounds more interesting than EPD. The default Python is 2.7.3, with numpy 1.5.1 (!) and scipy 0.11.0.

Bokeh There are a few dependencies to resolve before using Bokeh. This

includes Chaco and Traits, as well as Pandas but this one always compiled fine on my Mac.

I tried to compile Chaco last year; I don't remember if it was from a manual install or through pip, but it failed. Today it just works (such packages usually go to my system-wide library, /Library/Python/2.7/site-packages):

% sudo pip install Chaco

I should note that several packages were installed or updated at the same time, including pyface, traitsui, enable, and kiva.

Exemples from the (Quickstart)[http://docs.enthought.com/chaco/quickstart.html] guide seem to run smoothly. I had to replace traits.ui by traitsui, and of course remove every mention of enthought.*

To install Bokeh, I just followed the instructions on GitHub.

% git clone git://github.com/ContinuumIO/Bokeh.git
% cd Bokeh
% sudo python setup.py install

That worked just fine to generate static html on my disk (local_example.py). As I wanted to use the websockets backend, there're a couple of additional steps I had to follow. I already have redis installed on my Mac.

% python startlocal.py
Traceback (most recent call last):
  File "startlocal.py", line 1, in <module>
    from bokeh.server import start
  File "/Users/chl/gitroot/Bokeh/bokeh/server/start.py", line 1, in <module>
    from geventwebsocket.handler import WebSocketHandler
ImportError: No module named geventwebsocket.handler

So, I had to install gevent-websocket, and then flask. Finally, following the instructions on GitHub, this is just a matter of starting the redis server, launch python startlocal.py in Bokeh main directory, and run python pandas_example.py in another terminal to get the following plot, live in my browser (http://localhost:5006/bokeh):

Quite impressive so far!

Numba and Blaze

Other projects I started to look at are http://numba.pydata.org and Blaze which is presented as a generalization of NumPy.

Notes

(a) I should note that Enthought just released Canopy (or see Introducing Enthought Canopy), but this is probably another story, and I tend to keep away from multiple install of Python. I had some issues with EPD and IPython last year so I decided to only use the Python version that ships with my Mac. I know some folks recommend using an official Python package, but I never have had any problem with Apple Python.

---

Articles with the same tag(s):

Yet another gray theme for R base graphics
Light Table and interactive live coding
Bar charts of counts or frequencies in Stata
Emacs auto-completion for Python
Interactive data visualization with cranvas
Interacting with Weka from Jython
Easy creation of videos with R
Biplots
IPython 0.11, the new killer app
Installing numpy+scipy on OS X Lion

---