How to give sns.clustermap a precomputed distance matrix in Matplotlib?

To use sns.clustermap with a precomputed distance matrix in Matplotlib, you need to pass your distance matrix to the row_linkage and col_linkage parameters. This allows you to provide custom clustering results instead of letting seaborn compute distances automatically.

Creating a Basic Clustermap with Default Distance

First, let's see how clustermap() works with default distance calculation ?

import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from scipy.cluster.hierarchy import linkage
from scipy.spatial.distance import pdist

plt.rcParams["figure.figsize"] = [10, 6]
sns.set_theme(color_codes=True)

# Load example dataset
iris = sns.load_dataset("iris")
species = iris.pop("species")

# Create basic clustermap
g = sns.clustermap(iris, figsize=(8, 6))
plt.show()

Using Precomputed Distance Matrix

To use a precomputed distance matrix, calculate distances first and create linkage matrices ?

import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from scipy.cluster.hierarchy import linkage
from scipy.spatial.distance import pdist

# Load and prepare data
iris = sns.load_dataset("iris")
species = iris.pop("species")
iris_data = iris.values

# Compute distance matrices
row_distances = pdist(iris_data, metric='euclidean')
col_distances = pdist(iris_data.T, metric='euclidean')

# Create linkage matrices from precomputed distances
row_linkage = linkage(row_distances, method='ward')
col_linkage = linkage(col_distances, method='ward')

# Create clustermap with precomputed linkages
g = sns.clustermap(iris, 
                   row_linkage=row_linkage, 
                   col_linkage=col_linkage,
                   figsize=(8, 6))
plt.show()

Using Different Distance Metrics

You can use various distance metrics for clustering ?

import matplotlib.pyplot as plt
import seaborn as sns
from scipy.cluster.hierarchy import linkage
from scipy.spatial.distance import pdist

# Load data
iris = sns.load_dataset("iris")
species = iris.pop("species")

# Use Manhattan distance
row_distances = pdist(iris.values, metric='manhattan')
row_linkage = linkage(row_distances, method='complete')

# Create clustermap with Manhattan distance
g = sns.clustermap(iris, 
                   row_linkage=row_linkage,
                   col_linkage=None,  # Use default for columns
                   figsize=(8, 6),
                   cmap='viridis')
plt.show()

Key Parameters

Conclusion

Use precomputed distance matrices with sns.clustermap by passing linkage matrices to row_linkage and col_linkage parameters. This gives you full control over the clustering algorithm and distance metrics used.