Clustering Extensions

This page collects clustering tools that go beyond the basic Cluster, KMedoids, ClusterQuality, and ClusterResults workflow. Use these tools when you need to compare a range of cluster solutions, validate partitions, work with fuzzy memberships, or build property-based typologies.

At a Glance

Question	Guidance
Use this when	A single hard clustering solution is not enough: you need range search, fuzzy membership, validation, or property-based interpretation.
You need before starting	A distance matrix, a candidate partition, or a SequenceData object depending on the tool.
Do not use this when	You only need a first standard partition; start with KMedoids or Cluster.
Next step	Use Cluster Quality Indicators and the tool-specific result tables together.

Cluster Ranges and Method Comparison

Function or class	Role
k_medoids_range()	Run weighted PAM/k-medoids for several values of k
hierarchical_cluster_range()	Evaluate hierarchical clustering solutions across k
compare_cluster_methods()	Compare multiple hierarchical and PAM-style methods on one distance matrix
ClusterRangeResult	Result object for one family of partitions
ClusterRangeFamilyResult	Result object for multiple clustering methods

from sequenzo.clustering import compare_cluster_methods

comparison = compare_cluster_methods(
    diss,
    maxcluster=10,
    weights=weights,
    methods="all",
    random_state=42,
)

print(comparison.allstats)

Use this when your question is "which clustering method and number of clusters are plausible?" rather than "fit this one chosen clustering solution."

Partition Validation

Function	Purpose
compute_partition_quality()	Compute partition quality indicators for one clustering
cluster_range_from_partitions()	Evaluate a table of candidate partitions
boot_cluster_range()	Bootstrap partition quality over resamples
observation_silhouette()	Observation-level silhouette diagnostics
cluster_association()	Association between cluster labels and covariates
plot_cluster_association()	Visualize cluster-covariate association summaries
rarcat()	Typology regression validation following the RARCAT idea

from sequenzo.clustering import compute_partition_quality, observation_silhouette

quality = compute_partition_quality(diss, cluster_labels, weights=weights)
sil = observation_silhouette(diss, cluster_labels)

Fuzzy Clustering and Memberships

Function or class	Role
wfcmdd()	Weighted fuzzy clustering for distance data
WfcmddResult	Result object returned by wfcmdd()
crispness()	Crispness score for membership matrices
get_fuzzy_clusters()	Run fuzzy clustering through a unified FANNY or wfcmdd interface
membership_summary()	Summarize membership strength
most_typical_members()	Identify high-membership cases
fuzzy_sequence_plot()	Plot fuzzy sequence groups
dirichlet_regression()	Model fuzzy memberships with covariates
beta_regression()	Model one membership dimension

from sequenzo.clustering import get_fuzzy_clusters, membership_summary

fuzzy = get_fuzzy_clusters(
    diss,
    n_clusters=4,
    method="wfcmdd",
    weights=weights,
)
summary = membership_summary(fuzzy.membership)

Fuzzy clustering is useful when many cases sit between ideal trajectory types and hard cluster assignment hides that ambiguity.

Property-Based Clustering

Property-based clustering uses sequence-derived features to build interpretable typologies.

Function	Purpose
extract_sequence_properties()	Extract state, duration, transition, pattern, and complexity properties
property_based_clustering()	Build a discrepancy tree from extracted properties
seqpropclust()	WeightedCluster::seqpropclust()-compatible alias
cluster_split_schedule()	Order tree splits by global relevance
cut_tree()	Cut a property tree into groups
prune_property_tree()	Prune the tree to a simpler typology
tree_labels()	Generate readable labels for terminal groups
plot_property_tree()	Visualize the property tree
property_clustering_quality()	Evaluate property-clustering quality

from sequenzo.clustering import property_based_clustering, plot_property_tree

tree = property_based_clustering(
    seqdata,
    diss,
    properties=("state", "duration", "transition", "Complexity"),
    max_clusters=6,
)

plot_property_tree(tree)

Use this approach when interpretability of the splitting variables matters as much as distance-based compactness.

Relationship to Core Clustering Pages

• Start with Hierarchical Clustering or KMedoids for a standard partition.
• Use Cluster Quality to inspect one selected solution.
• Use this page when you need range search, method comparison, fuzzy memberships, bootstrap validation, or property-based typologies.

Authors

Code: Yuqi Liang