基于DBSCAN的轨迹路径识别报错求助：TypeError问题排查

Let's walk through exactly why this error is popping up and fix your DBSCAN clustering code for trajectory data.

The Root Cause #

That error is telling you that np.radians() is being handed individual float values instead of a structured array it can iterate over. Let's look at how you're prepping your coordinates:

coords = a['Long']
coords['Lat'] = a['Lat']
coords = coords.to_numpy(coords)

Here, you start with a 1D Series of longitude values, then try to tack on latitude values as a new "key" in that Series—this doesn't create the 2D coordinate pairs DBSCAN needs. Instead, you end up with a messy 1D structure, and when np.radians() tries to process it, it's looking for a .radians() method on each float (which doesn't exist).

The Fixed Code #

Here's the corrected version of your function, with explanations for each change:

import numpy as np
from sklearn.cluster import DBSCAN

def clustersDBSCAN(data):
    # Make a copy of the input to avoid modifying the original DataFrame
    trajectory_data = data.copy()
    
    # Create a proper 2D array of [Lat, Long] pairs (critical for haversine metric!)
    # Note: Haversine expects coordinates in (latitude, longitude) order
    coords = trajectory_data[['Lat', 'Long']].to_numpy()
    
    kms_per_radian = 6371.0088
    # Convert 0.02 km to radians for the epsilon threshold
    epsilon = 0.02 / kms_per_radian
    
    # Fit DBSCAN with radians-converted coordinates
    db = DBSCAN(
        eps=epsilon,
        min_samples=1,
        algorithm='ball_tree',
        metric='haversine'
    ).fit(np.radians(coords))
    
    # Attach cluster labels to the DataFrame
    trajectory_data['clusters'] = db.labels_
    return trajectory_data

Key Fixes Explained #

• Proper 2D Coordinate Array: By using trajectory_data[['Lat', 'Long']].to_numpy(), we create a 2D array where each row is a valid (latitude, longitude) pair. This is exactly what np.radians() and the haversine metric expect.
• Order Matters: The haversine metric in scikit-learn requires coordinates in (lat, lon) order—swapping these would lead to incorrect distance calculations, even if the error goes away.
• Avoid Modifying Original Data: Using data.copy() prevents accidental changes to your input DataFrame, which is a safe practice for data processing functions.
• Correct to_numpy() Usage: The original code passed coords as an argument to to_numpy(), which is unnecessary—to_numpy() doesn't need that parameter.

Quick Pre-Run Check #

Make sure your Lat and Long columns in the input DataFrame are numeric (float or int) types. If they're stored as strings, convert them first with:

import pandas as pd

data['Lat'] = pd.to_numeric(data['Lat'], errors='coerce')
data['Long'] = pd.to_numeric(data['Long'], errors='coerce')

内容的提问来源于stack exchange，提问作者nicolax9777