Affinity Propagation

Clustering by passing messages between data points. Automatically determines the number of clusters.

When to use:

• Don't know number of clusters
• All data points are potential cluster centers
• Small to medium datasets
• Want exemplar-based clustering

Strengths: Automatically finds k, each cluster has an exemplar, no initialization needed, deterministic with fixed random state Weaknesses: Very slow (O(n²) memory, O(n²×iterations) time), many parameters to tune, may not converge

Model Parameters

Damping (default: 0.5) Controls message update stability (0.5 to 1.0).

• 0.5-0.7: More responsive, may oscillate
• 0.7-0.9: More stable
• 0.9-1.0: Very stable but slow convergence

Max Iterations (default: 200) Maximum number of message-passing iterations.

• 100-200: Usually sufficient
• 500+: For difficult convergence

Convergence Iterations (default: 15) Number of iterations with no change that indicates convergence.

• 10-15: Standard
• 20+: More conservative

Random State (default: 42) Seed for reproducibility.