Hierarchical Clustering

Builds a tree (dendrogram) of clusters by iteratively merging or splitting groups based on distance.

When to use:

• Want to visualize cluster hierarchy with dendrogram
• Need clusters at multiple granularities
• Relatively small dataset (<10k samples)
• Want deterministic results

Strengths: Creates hierarchical structure, no need to specify k upfront, deterministic, visualizable with dendrogram Weaknesses: Slow on large datasets, sensitive to noise and outliers, cannot undo merges

Model Parameters

N Clusters (default: 2, required) Number of clusters to extract from the hierarchy.

Linkage (default: "ward") How to measure distance between clusters:

• ward: Minimizes variance (default, best for most cases)
• complete: Maximum distance between all point pairs
• average: Average distance between all point pairs
• single: Minimum distance (can create long chains)

Metric (default: "euclidean") Distance metric for computing linkage:

• euclidean: Standard distance (required for ward)
• manhattan: L1 distance
• cosine: Angle-based similarity
• Others: l1, l2, correlation, etc.

Distance Threshold (optional) Stop merging when distance exceeds this threshold. If set, n_clusters should be None.

Compute Full Tree (default: "auto") Whether to compute the full tree or stop early:

• auto: Automatically decide based on parameters
• true: Compute full dendrogram
• false: Stop early (faster)