Affinity propagation parameters

• Edge weight attribute (click refresh if no attribute names are listed) - this attribute provides the similarities between datapoints: Affinity Propagation takes as input a collection of real-valued similarities between data points, where the similarity s(i; k) indicates how well the data point with index k is suited to be the exemplar for data point i.
• Number of iterations - maximum number of iterations to be performed
• Stop criterion (n) - algorithm will stop after the n-th iteration without any change in centers (exemplars). In this situation, we say that the algorithm converged. If the maximum number of interactions is reached before the stop criterion is satisfied, then the algorithm did not converge. The current clusters can still be retrieved in this situation.
• Preference - the preference for a node to be an exemplar (cluster center); currently equal for every node.
• Lambda (damping factor) - when updating the messages, each message is set to λ times its value from the previous iteration plus 1 - λ times its prescribed updated value, where the damping factor λ is between 0 and 1.
• Noise option - for some networks adding noise is crucial to arrive at meaningful clusters. One example is in situations where two or more nodes have equal affinity to become exemplars. In such cases adding noise breaks the symmetry and allows one of the nodes to be selected. The original implementation adds noise to the similarity values associated with edges. We found that more stable results are achieved when noise is added to node preferences. Currently noise is equal to preference* epsilon * (random real number from a uniform distribution on [0; 1]) where epsilon = 0.0000001.

Other implementation-specific parameters

• Cluster ID - name of the node attribute which will store cluster assignments
• Center ID - name of the node attribute which will store center assignments
• Take log - if selected, edges weights will be set to log(edge_weight)