triangular_clustermap.py

import string

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

# generate some data
n_groups = 4
n_per_group = 5
n = n_groups * n_per_group
labels = pd.Categorical([x for x in string.ascii_uppercase[:n_groups] for y in range(n_per_group)], ordered=True)
colors = plt.get_cmap('tab10')(labels.codes)
x = pd.DataFrame(pd.get_dummies(labels).values + np.random.random((n, n_groups)), index=labels)

c = x.T.corr()

kws = dict(row_colors=colors, dendrogram_ratio=0.08, cbar_kws=dict(label='correlation'), figsize=(6.1, 6))

# won't work because 'c' is not guaranteed to be ordered by dendrogram order
grid = sns.clustermap(c, mask=np.triu(c) == 0, **kws)

# Option 1:
# # first cluster and get order
grid1 = sns.clustermap(c)
order = grid1.dendrogram_row.reordered_ind
co = c.iloc[order, order]
plt.close(grid1.fig)
# # then, mask lowe halt
grid2 = sns.clustermap(co, mask=np.triu(co) == 0, xticklabels=False, row_cluster=False, **kws)
# # rotate yaxislabels
grid2.ax_heatmap.set_yticklabels(grid2.ax_heatmap.get_yticklabels(), rotation=0)


# Option 2:
# # plot
grid = sns.clustermap(c, xticklabels=False, **kws)
# # remove lower halt, by setting values to nan
mesh = grid.ax_heatmap.get_children()[0]
c[np.triu(c) == 0] = np.nan
mesh.set_array(c)
# # remove row dendrogram
grid.fig.delaxes(grid.ax_row_dendrogram)
# # rotate yaxislabels
grid.ax_heatmap.set_yticklabels(grid.ax_heatmap.get_yticklabels(), rotation=0)