robroc · April 29, 2024 11:35
diff --git a/hdbscan_map_clusters.py b/hdbscan_map_clusters.py
 # Thanks to Leland McInnes (@leland_mcinnes) for devising the weighting solution

 import hdbscan
 # https://hdbscan.readthedocs.io/en/latest/index.html

 import numpy as np

 """Assume df is a pandas dataframe with latitude and longitude columns in WGS84"""

 # Define a cluster selection epsilon to group points within this distance.
 # In this case, it's 10 meters (0.01 km divided by the radius of the Earth).
 eps = .01 / 6378

 # Initiate HDBSCAN clusterer
 clusterer = hdbscan.HDBSCAN(min_cluster_size = 200,
                            metric = 'haversine',
                            cluster_selection_epsilon = eps)

 # Convert coordinates to radians for distance-based grouping and cluster them
 clusterer.fit(np.radians(df[['latitude', 'longitude']]).values)

 # Assign cluster labels to each point
 df['clusters'] = clusterer.labels_

 # Assign the membership strenth to each point. The closer a point is the a high-density area, the stronger the membership.
 # This will be the weighting factor.
 df['membership_strength'] = clusterer.probabilities_

 # Define a weighted average function with numpy.
 weighted_mean = lambda x: np.average(x, weights=df.loc[x.index, 'membership_strength'])

 # Exclude points with label -1. These are points that weren't assigned to a cluster and have membership strength of 0.
 # Then group them by cluster labels and use the weighted average of the coordinates to find the cluster center.
 clustered_zones = (df.query('clusters != -1').groupby('clusters')
                     .agg(dict(cluster_size = 'sum', 
                         longitude = weighted_mean, 
                         latitude = weighted_mean))
    )
	# Thanks to Leland McInnes (@leland_mcinnes) for devising the weighting solution

	import hdbscan
	# https://hdbscan.readthedocs.io/en/latest/index.html

	import numpy as np

	"""Assume df is a pandas dataframe with latitude and longitude columns in WGS84"""

	# Define a cluster selection epsilon to group points within this distance.
	# In this case, it's 10 meters (0.01 km divided by the radius of the Earth).
	eps = .01 / 6378

	# Initiate HDBSCAN clusterer
	clusterer = hdbscan.HDBSCAN(min_cluster_size = 200,
	metric = 'haversine',
	cluster_selection_epsilon = eps)

	# Convert coordinates to radians for distance-based grouping and cluster them
	clusterer.fit(np.radians(df[['latitude', 'longitude']]).values)

	# Assign cluster labels to each point
	df['clusters'] = clusterer.labels_

	# Assign the membership strenth to each point. The closer a point is the a high-density area, the stronger the membership.
	# This will be the weighting factor.
	df['membership_strength'] = clusterer.probabilities_

	# Define a weighted average function with numpy.
	weighted_mean = lambda x: np.average(x, weights=df.loc[x.index, 'membership_strength'])

	# Exclude points with label -1. These are points that weren't assigned to a cluster and have membership strength of 0.
	# Then group them by cluster labels and use the weighted average of the coordinates to find the cluster center.
	clustered_zones = (df.query('clusters != -1').groupby('clusters')
	.agg(dict(cluster_size = 'sum',
	longitude = weighted_mean,
	latitude = weighted_mean))
	)