Clustering

class igua.clustering.ClusteringStrategy

An abstract clustering strategy to cluster compositional data.

abstract cluster(X, weights=None)

Cluster the given observations.

Parameters:

• X (scipy.sparse.csr_matrix) – A matrix of shape \(m \times n\) with compositional data.

• weights (numpy.ndarray or None) – The weights (of shape \(n\)) to use for computing distances. If None, use uniform weights.

Returns:

numpy.ndarray – A flat array of shape \(m\) which assigns an arbitrary cluster number to each observation of X (see scipy.cluster.hierarchy.fcluster documentation).

class igua.clustering.HierarchicalClustering(ClusteringStrategy)

A clustering strategy implementing hierarchical clustering.

__init__(*, method, distance=0.8, precision='double', jobs=1)

Create a new hierarchical clustering strategy.

Parameters:

• method (str) – The name of the linkage method to use: average, single, complete, weighted, centroid, median or ward.

• distance (float) – The distance cutoff for the flat clusters created with scipy.cluster.hierarchy.fcluster.

• precision (str) – The floating-point precision to use: half, single or double. Note that changing precision (in particular with half-precision) may lead to differences in the generated clustering.

• jobs (int) – The number of parallel threads to use to perform the pairwise distance computation.

cluster(X, weights=None)

Cluster the given observations.

Parameters:

• X (scipy.sparse.csr_matrix) – A matrix of shape \(m \times n\) with compositional data.

• weights (numpy.ndarray or None) – The weights (of shape \(n\)) to use for computing distances. If None, use uniform weights.

Returns:

numpy.ndarray – A flat array of shape \(m\) which assigns an arbitrary cluster number to each observation of X (see scipy.cluster.hierarchy.fcluster documentation).

class igua.clustering.LinearClustering(ClusteringStrategy)

A clustering strategy similar to MMseqs2 linear clustering.

__init__(*, distance=0.8)

Create a new linear clustering strategy.

Parameters:

distance (float) – The distance cutoff to use for clustering observations together.

cluster(X, weights=None)

Cluster the given observations.

Parameters:

• X (scipy.sparse.csr_matrix) – A matrix of shape \(m \times n\) with compositional data.

• weights (numpy.ndarray or None) – The weights (of shape \(n\)) to use for computing distances. If None, use uniform weights.

Returns:

numpy.ndarray – A flat array of shape \(m\) which assigns an arbitrary cluster number to each observation of X (see scipy.cluster.hierarchy.fcluster documentation).