clusterer

A clusterer is a computational method or model that assigns objects to groups, or clusters, such that objects within the same cluster are more similar to each other than to objects in other clusters. In machine learning and data mining, clusterers operate in an unsupervised setting, using only feature data and no labeled outcomes. The result is a clustering: for each data point, a cluster label is produced, and many algorithms also provide additional information such as cluster centroids, compactness measures, or membership probabilities.

Clusterers can be broadly categorized as hard or soft. Hard clusterers assign each object to a single cluster, while soft clusterers produce probabilities or degrees of membership across clusters. Common families include centroid-based methods (such as k-means and k-medoids), hierarchical methods (agglomerative and divisive), density-based methods (DBSCAN, OPTICS), grid-based methods, and model-based methods (Gaussian mixture models, expectation-maximization). Distance metrics and feature preparation influence results; most clusterers assume numeric features and often require scaling, while handling categorical features may require encoding or specialized distance measures.

Output from a clusterer typically includes cluster assignments and, depending on the algorithm, additional information such as cluster centers, size, and membership scores. Practical considerations include scalability to large datasets, sensitivity to initialization and outliers, and the need to select parameters (for example, the number of clusters k or density thresholds). Evaluation uses internal indices (like silhouette score) or, when available, external indices comparing clusters with ground-truth labels.