如何用Python的DBSCAN算法对给定3D点坐标进行聚类？

Hey there! As a programming newb looking to use DBSCAN for 3D point clustering in Python, I’ve got you covered with a complete, easy-to-follow example. Let’s break this down step by step.

First, you’ll need a few key packages. Scikit-learn has a ready-to-use DBSCAN implementation, numpy handles numerical data, and matplotlib lets you visualize your 3D clusters. Install them with this command:

pip install scikit-learn numpy matplotlib

Let’s take your example points and turn them into a format DBSCAN understands. We’ll use a numpy array where each row represents a 3D coordinate:

import numpy as np

# Example 3D points (add as many as you need)
points = np.array([
    [-37.530, 3.109, -16.452],
    [40.247, 5.483, -15.209],
    [-36.890, 2.987, -16.123],
    [39.567, 5.123, -14.890],
    [-38.120, 3.210, -16.678],
    [41.012, 5.678, -15.567],
    [10.000, 2.000, 5.000],  # This will be noise if eps is small
])

Now, let’s set up and run the DBSCAN algorithm. The two most important parameters are:

• eps: The maximum distance between two points for them to be considered part of the same neighborhood.
• min_samples: The minimum number of points required to form a cluster (including the point itself).

For your 3D data, you’ll need to adjust eps based on how spread out your points are. Let’s start with a reasonable value for the example:

from sklearn.cluster import DBSCAN

# Initialize DBSCAN
dbscan = DBSCAN(eps=2.0, min_samples=2)

# Fit the model to your points
clusters = dbscan.fit_predict(points)

The clusters array gives a label for each point. A label of -1 means the point is classified as noise (doesn’t belong to any cluster). Let’s print out what we got:

# Number of clusters (excluding noise)
num_clusters = len(set(clusters)) - (1 if -1 in clusters else 0)
num_noise = list(clusters).count(-1)

print(f"Number of clusters: {num_clusters}")
print(f"Number of noise points: {num_noise}")
print(f"Cluster labels for each point: {clusters}")

For our example, you should see two clusters (the first three and next three points) and one noise point (the last one).

Seeing the clusters in 3D helps make sense of the results. Let’s use matplotlib to plot them:

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

# Create a 3D plot
fig = plt.figure(figsize=(10, 8))
ax = fig.add_subplot(111, projection='3d')

# Plot each cluster with a different color
unique_labels = set(clusters)
colors = [plt.cm.Spectral(each) for each in np.linspace(0, 1, len(unique_labels))]

for label, color in zip(unique_labels, colors):
    if label == -1:
        # Noise points are gray
        color = [0.5, 0.5, 0.5, 1]
    
    # Get indices of points in this cluster
    idx = clusters == label
    ax.scatter(points[idx, 0], points[idx, 1], points[idx, 2], c=[color], label=f"Cluster {label}" if label != -1 else "Noise")

ax.set_xlabel('X Coordinate')
ax.set_ylabel('Y Coordinate')
ax.set_zlabel('Z Coordinate')
ax.legend()
plt.show()

• Choosing eps: If your clusters are too small or too many, try increasing eps. If clusters merge when they shouldn’t, decrease it. For beginners, you can test a few values (like 1.0, 2.0, 5.0) to see what works.
• Choosing min_samples: If you have noisy data, increase this value to filter out more noise. If you have small clusters, decrease it.
• Scaling your data: If your coordinates are on very different scales (e.g., X ranges from -100 to 100, Y from 0 to 1), use StandardScaler from scikit-learn to normalize the data first—this helps eps work consistently across all dimensions.

Hope this helps you get started! If you need help tuning parameters for your specific dataset, feel free to share more details.

内容的提问来源于stack exchange，提问作者bob