Hierarchical clustering of heatmap in python

Question

I have a NxM matri with values that range from 0 to 20. I easily get an heatmap by using Matplotlib and pcolor. Now I'd like to apply a hierarchical clustering and a dendogram using scipy. I'd like to re-order each dimension (rows and columns) in order to show which element are similar (according to the clustering result). If the matrix would be square (NxN) the code would be something like:

clustering = linkage(matrix, method="average")
dendrogram(clustering, orientation='right')

How can I obtain the linkage matrix when the dimensions are different? Do I have to calculate manually each distance between every pair of elements?

Solution

I'm not sure how to do it using matplotlib and pcolor. However, seaborn has native support for what you're trying to accomplish.

http://seaborn.pydata.org/generated/seaborn.clustermap.html

Example:

import seaborn as sns; sns.set()
flights = sns.load_dataset("flights")
flights = flights.pivot("month", "year", "passengers")
g = sns.clustermap(flights)

Comments

[PlasmaBinturong]

Good to know! However its computing the similarity itself, what if I already have a similarity matrix to provide (e.g a covariance matrix from a PCA)?