Summary  
This chapter covers generating a clustermap by applying hierarchical clustering to matrix data and visualizing the result with dendrograms. It explains how to adjust clustering through parameters like data scaling, distance metrics, and linkage methods.

General domain of usage  
Exploratory data analysis

A `clustermap` is a matrix plot that combines a **heatmap** with **hierarchical clustering**.

While a standard heatmap displays data in a fixed grid, a clustermap **reorders** the rows and columns to place similar values next to each other. The tree-like diagrams on the axes are called **dendrograms**, and they show how the data points are grouped.



## Key Parameters

To control how the clustering works, you can use these parameters:

* **`standard_scale`**: standardizes the data (0 for rows, 1 for columns) so that each feature has a mean of 0 and variance of 1. This is crucial when variables have different units;
* **`metric`**: the distance measure to use (e.g., `'euclidean'`, `'correlation'`). It determines what "similar" means;
* **`method`**: the linkage algorithm to use (e.g., `'single'`, `'complete'`, `'average'`). It determines how to group the clusters.



## Example

Here is a clustermap of the Iris dataset. Notice how the species (rows) are automatically grouped together because they have similar measurements.

import seaborn as sns
import matplotlib.pyplot as plt

# Load dataset
df = sns.load_dataset('iris')
# Prepare matrix (drop non-numeric column for calculation)
species = df.pop("species")

# Create a clustermap
sns.clustermap(
    data=df,
    standard_scale=1,    # Normalize columns
    metric='euclidean',  # Measure distance
    method='average',    # clustering method
    cmap='viridis',
    figsize=(6, 6)
)

plt.show()

import unittest
import importlib
import sys
import pandas as pd
from unittest.mock import patch, MagicMock

# Helper function to dynamically generate test names and assertions
def _dynamic_test(test_case, condition, success_message, failure_message):
    if condition:
        test_case._testMethodName = success_message
        test_case.assertTrue(True, success_message)
    else:
        test_case._testMethodName = failure_message
        test_case.fail(failure_message)

class TestUserCode(unittest.TestCase):

    def setUp(self):
        # Mocking read_csv to prevent network issues and ensure consistent data
        self.patcher_csv = patch('pandas.read_csv')
        self.mock_read_csv = self.patcher_csv.start()
        
        # Create a simple DataFrame suitable for pivoting
        self.mock_df = pd.DataFrame({
            'year': [1949, 1949, 1950, 1950],
            'month': ['Jan', 'Feb', 'Jan', 'Feb'],
            'passengers': [112, 118, 115, 126]
        })
        self.mock_read_csv.return_value = self.mock_df

    def tearDown(self):
        self.patcher_csv.stop()

    # Test custom style configuration
    def test_style_configuration(self):
        with patch('seaborn.clustermap'), patch('matplotlib.pyplot.show'):
            with patch('seaborn.set_style') as mock_style:
                if 'user_code' in sys.modules:
                    importlib.reload(sys.modules['user_code'])
                import user_code
            
            if not mock_style.called:
                 _dynamic_test(self, False, "", "Expected `sns.set_style()` to be called.")
                 return

            args, kwargs = mock_style.call_args
            style_name_correct = args[0] == 'ticks'
            
            # Check dictionary configuration for facecolor
            rc_params = args[1] if len(args) > 1 else kwargs.get('rc', {})
            facecolor_correct = rc_params.get('figure.facecolor') == 'seagreen'

            _dynamic_test(
                self,
                style_name_correct and facecolor_correct,
                "The style is set to 'ticks' with 'seagreen' background.",
                "Expected `sns.set_style('ticks', {'figure.facecolor': 'seagreen'})`."
            )

    # Test clustermap parameters
    def test_clustermap_params(self):
        with patch('seaborn.clustermap') as mock_cluster:
            with patch('matplotlib.pyplot.show'):
                if 'user_code' in sys.modules:
                    importlib.reload(sys.modules['user_code'])
                import user_code
            
            if not mock_cluster.called:
                _dynamic_test(self, False, "", "Expected `sns.clustermap()` to be used.")
                return

            args, kwargs = mock_cluster.call_args
            
            # 1. Check Data Binding
            # Since user_code pivots the data, we check if the first arg is the pivoted dataframe
            passed_data = args[0] if args else kwargs.get('data')
            expected_data = user_code.upd_df
            data_check = passed_data is expected_data

            # 2. Check Parameters
            cmap_check = kwargs.get('cmap') == 'vlag'
            scale_check = kwargs.get('standard_scale') == 1
            method_check = kwargs.get('method') == 'single'
            metric_check = kwargs.get('metric') == 'correlation'
            annot_check = kwargs.get('annot') is True
            vmin_check = kwargs.get('vmin') == 0
            vmax_check = kwargs.get('vmax') == 10

            _dynamic_test(
                self,
                all([data_check, cmap_check, scale_check, method_check, metric_check, annot_check, vmin_check, vmax_check]),
                "The `clustermap` parameters (`standard_scale`, `method`, `metric`, `vmin`, etc.) are configured correctly.",
                f"Expected `standard_scale=1`, `method='single'`, `metric='correlation'`, `vmin=0`. Got: {kwargs}"
            )

    # Test show
    def test_show_used(self):
        with patch('seaborn.clustermap'):
            with patch('matplotlib.pyplot.show') as mock_show:
                if 'user_code' in sys.modules:
                    importlib.reload(sys.modules['user_code'])
                import user_code

                _dynamic_test(
                    self,
                    mock_show.called,
                    "The `plt.show()` function is used.",
                    "Expected `plt.show()` to be used to display the plot."
                )

if __name__ == '__main__':
    unittest.main()

test_code.py

Explore the power of statistical data visualization designed to reveal patterns and relationships. Use Seaborn to create informative distribution plots, heatmaps, and categorical charts with minimal code. Learn to apply aesthetic themes and color palettes that make complex statistical data accessible and easy to interpret.

Performing Hierarchical Clustering

Key Parameters

Example

解答