milia.cluster.consensus#

Consensus (ensemble) clustering through a co-association matrix (Fred and Jain, 2005).

Several clusterings of the same rows vote on whether each pair belongs together; the share of votes is a similarity, and average-linkage clustering of 1 - share gives the consensus partition. Pairs that the members disagree on end up between clusters rather than forced into one.

Functions

coassociation(labelings)

Share of labellings that put each pair of rows in the same (non-noise) cluster.

Classes

Consensus([estimators, n_clusters])

Consensus of several clusterers, or of one clusterer over several seeds.

milia.cluster.consensus.coassociation(labelings)#

Share of labellings that put each pair of rows in the same (non-noise) cluster.

Parameters:

labelings (list[ndarray]) – Several labellings of the same rows.

Return type:

ndarray

Returns:

Square matrix of shares, 1 on the diagonal for non-noise rows.

Examples

>>> coassociation([np.array([0, 0, 1]), np.array([0, 1, 1])])[0].tolist()
[1.0, 0.5, 0.0]
class milia.cluster.consensus.Consensus(estimators=None, n_clusters=3)#

Bases: BaseClusterer

Consensus of several clusterers, or of one clusterer over several seeds.

Parameters:
  • estimators (list[BaseEstimator] | None) – Member clusterers; k-means with five seeds by default.

  • n_clusters (int) – Clusters in the consensus partition (also set on members that take n_clusters).

Variables:

coassociation – The co-association matrix of the training rows.

Examples

>>> rng = np.random.default_rng(0)
>>> X = np.repeat(np.array([[0.0, 0.0], [6.0, 6.0], [0.0, 6.0]]), 20, axis=0) + rng.normal(
...     scale=0.4, size=(60, 2)
... )
>>> model = Consensus(n_clusters=3).fit(X)
>>> model.n_clusters_, model.coassociation_.shape
(3, (60, 60))