mpl_sankey

This is a simple package to produce sankey plots with matplotlib. It is an alternative to the matplotlib.sankey module, which sacrifices some flexibility in exchange for a much simpler interface.

Specifically, this package only produces standard horizontal sankey plots, but it does so automatically from data, rather than requiring the user to figure out the spatial distribution of nodes/flows.

How it works

There is a single method, sankey(), which accepts one argument data and several optional keyword arguments.

data must be a 2-dimensional tabular object with at least 3 columns: best results are obtained with pandas.DataFrames, but numpy.arrays and simple lists of lists are also accepted.

• each row corresponds to a "flow"
• the first column stores the weights (that is, the tickness of each flow drawn), and must be an int or float
• each other column denotes a node (by its label, which can have any type) where the flow passes, at each stage
• if data is a pandas.DataFrame, then the column names are represented on top of each stage

Example

from mpl_sankey import sankey
from matplotlib import pyplot as plt
import pandas as pd

data = pd.DataFrame([[1, 'a', 1, 'I', 1, 'success'],
                     [2, 'b', 2, 'III', 2, 'discard'],
                     [1, 'b', 1, 'II', 2, 'success'],
                     [1, 'c', 1, 'II', 2, 'discard'],
                     [2.5, 'a', 2, 'IV', 1, 'discard'],
                     [2, 'a', 1, 'I', 1, 'success']],
                    columns=['Weight', 'First', 'Then', 'After', 'Finally', 'Outcome'])
                    
plt.figure(figsize=(12, 3))
sankey(data, cmap=plt.get_cmap('viridis'))
plt.savefig('featured.png', bbox_inches='tight')

Find more in the Examples notebook.

Installation and dependencies

You can install this package through PyPi with pip install mpl_sankey, or just clone the repo from github.

The package requires pandas (and, obviously, matplotlib) to work.