dbscan-custom-distance

name	dbscan-custom-distance
description	Define custom distance metrics for DBSCAN clustering. Use this skill whenever a task requires clustering points with unequal weighting or application-specific distances.

DBSCAN Custom Distance Metrics

This skill shows how to define and use custom distance metrics with scikit-learn's DBSCAN.

Parameterized Distance Functions

To use a distance function with configurable parameters (like variable feature weights), use a factory function:

import numpy as np
from sklearn.cluster import DBSCAN

def create_weighted_distance(weight_x, weight_y):
    """Create a distance function with specific weights."""
    def distance(a, b):
        dx = a[0] - b[0]
        dy = a[1] - b[1]
        return np.sqrt((weight_x * dx)**2 + (weight_y * dy)**2)
    return distance

# Create a distance function instance
dist_metric = create_weighted_distance(1.5, 0.5)

# Pass it to DBSCAN
db = DBSCAN(eps=10, min_samples=5, metric=dist_metric)
labels = db.fit_predict(points)

Performance Notes

• Custom Python metrics are slower than built-in scikit-learn metrics.
• Vectorized operations using scipy.spatial.distance.pdist or cdist are faster, but DBSCAN handles the metric parameter elegantly via a callable.