active-constrained-clustering

Active constrained clustering algorithms for scikit-learn.

Algorithms

Semi-supervised clustering

• Seeded K-Means
• Constrained K-Means

Constrained clustering

• Metric K-Means (MK-Means)
• Pairwise constrained K-Means (PCK-Means)
• Constrained K-Means (COPK-Means)
• Metric pairwise constrained K-Means (MPCK-Means)

New additions:

• Constrained spectral clustering (CSP)
• Agglomerative hierarchical clustering with constraints (AHCC)

Active learning of pairwise clustering

• Explore-Consolidate
• Min-Max
• Normalized point-based uncertainty (NPU)

New additions:

• Borderline MST heuristic
• Distant MST heuristic
• Unified borderline and distant MST heuristics
• Explore-NPU

Installation

sudo apt-get update
sudo apt-get install git
pip install --upgrade git:git://github.com/heriosousa/active-constrained-clustering.git

Usage

from sklearn import datasets, metrics
from active_semi_clustering.semi_supervised.pairwise_constraints import PCKMeans
from active_semi_clustering.active.pairwise_constraints import ExampleOracle, ExploreConsolidate, MinMax

X, y = datasets.load_iris(return_X_y=True)

First, obtain some pairwise constraints from an oracle.

# TODO implement your own oracle that will, for example, query a domain expert via GUI or CLI
oracle = ExampleOracle(y, max_queries_cnt=10)

active_learner = MinMax(n_clusters=3)
active_learner.fit(X, oracle=oracle)
pairwise_constraints = active_learner.pairwise_constraints_

Then, use the constraints to do the clustering.

clusterer = PCKMeans(n_clusters=3)
clusterer.fit(X, ml=pairwise_constraints[0], cl=pairwise_constraints[1])

Evaluate the clustering using Adjusted Rand Score.

metrics.adjusted_rand_score(y, clusterer.labels_)