I am using HDBSCAN to generate prediction data for a given cluster model. I then attempt to classify new points using the approximate_predict function to find the correct cluster for a new point. The model returns the correct cluster for a new point but the probability/strength is always 0.0. To generate the model and classify new points I use:
# Generate the model
cluster_model = hdbscan.HDBSCAN(metric='euclidean', min_cluster_size=3, cluster_selection_epsilon=0.4,
prediction_data=True).fit(data)
# Classify new point
cluster, prob = hdbscan.approximate_predict(cluster_model, new_point)
print(cluster, prob)
As I understand, if the strength/probability is 0.0 the point has been classified as noise. From manual analysis of the new points I can see they do fit into the heart of the original clusters so I don't understand why the probability is always 0.0?