Dendrogram generated by scipy-cluster does not show


Question

I am using scipy-cluster to generate a hierarchical clustering on some data. As a final step of the application, I call the dendrogram function to plot the clustering. I am running on Mac OS X Snow Leopard using the built-in Python 2.6.1 and this matplotlib package. The program runs fine, but at the end the Rocket Ship icon (as I understand, this is the launcher for GUI applications in python) shows up and vanishes immediately without doing anything. Nothing is shown. If I add a 'raw_input' after the call, it just bounces up and down in the dock forever. If I run a simple sample application for matplotlib from the terminal it runs fine. Does anyone have any experiences on this?

1
13
6/3/2010 4:21:44 PM

Accepted Answer

I had the same issue on Ubuntu 10.04. In order to get graphics to display from ipython interactive console, start it with "-pylab" switch, which enables the interactive use of matplotlib:

ipython -pylab

To get your graphics to display during the execution of a standalone script, use matplotlib.pyplot.show call. Here's an example from hcluster homepage, the first and last line are the significant bits here:

from matplotlib.pyplot import show

from hcluster import pdist, linkage, dendrogram
import numpy
from numpy.random import rand

X = rand(10,100)
X[0:5,:] *= 2
Y = pdist(X)
Z = linkage(Y)
dendrogram(Z)

show()
16
9/28/2010 10:34:50 AM

Invoking ipython with "-pylab" switch didn't make a difference for me. (System: Fedora 13)

Though not ideal, my solution was to explicitly write the resulting figure as a file. For example:

...
dendrogram(Z)
pylab.savefig( "temp.png" )

Hope this helps anyone who is running into the same issue.

Amendment: Be careful about simply using copy-and-paste with the hcluster package's brief tutorial, notably in that if you call pylab.savefig() after several types of dendrogram drawing shown in the tutorial, i.e.

distMat = # whatever distance matrix you have
dendrogram( linkage( distMat ) )
pylab.savefig( "exampleDendrogram.png" )
dendrogram( linkage( distMat, method="complete" ) ) #instead of default "single"
pylab.savefig( "exampleDendrogram.png" )

Then exampleDendrogram.png will contain both the single-linkage dendrogram and the complete-linkage dendrogram in the same figure, and they will likely cross-cross and look like a mess.

If you're as stupid as me, you'll spend 30-180 minutes in confusion about how to properly use hcluster, when it's actually just a matter of resetting matplotlib between dendrogram calls:

distMat = # whatever distance matrix you have
dendrogram( linkage( distMat ) )
pylab.savefig( "exampleDendrogram1.png" )
pylab.cla()
dendrogram( linkage( distMat, method="complete" ) ) #instead of default "single"
pylab.savefig( "exampleDendrogram2.png" )

Now, the resulting dendrogram image files will look like what you expected them to look like.


Licensed under: CC-BY-SA with attribution
Not affiliated with: Stack Overflow
Icon