"""Bisecting K-means clustering.""" # Author: Michal Krawczyk import warnings import numpy as np import scipy.sparse as sp from ._kmeans import _BaseKMeans from ._kmeans import _kmeans_single_elkan from ._kmeans import _kmeans_single_lloyd from ._kmeans import _labels_inertia_threadpool_limit from ._k_means_common import _inertia_dense from ._k_means_common import _inertia_sparse from ..utils.extmath import row_norms from ..utils._openmp_helpers import _openmp_effective_n_threads from ..utils.validation import check_is_fitted from ..utils.validation import _check_sample_weight from ..utils.validation import check_random_state from ..utils._param_validation import StrOptions class _BisectingTree: """Tree structure representing the hierarchical clusters of BisectingKMeans.""" def __init__(self, center, indices, score): """Create a new cluster node in the tree. The node holds the center of this cluster and the indices of the data points that belong to it. """ self.center = center self.indices = indices self.score = score self.left = None self.right = None def split(self, labels, centers, scores): """Split the cluster node into two subclusters.""" self.left = _BisectingTree( indices=self.indices[labels == 0], center=centers[0], score=scores[0] ) self.right = _BisectingTree( indices=self.indices[labels == 1], center=centers[1], score=scores[1] ) # reset the indices attribute to save memory self.indices = None def get_cluster_to_bisect(self): """Return the cluster node to bisect next. It's based on the score of the cluster, which can be either the number of data points assigned to that cluster or the inertia of that cluster (see `bisecting_strategy` for details). """ max_score = None for cluster_leaf in self.iter_leaves(): if max_score is None or cluster_leaf.score > max_score: max_score = cluster_leaf.score best_cluster_leaf = cluster_leaf return best_cluster_leaf def iter_leaves(self): """Iterate over all the cluster leaves in the tree.""" if self.left is None: yield self else: yield from self.left.iter_leaves() yield from self.right.iter_leaves() class BisectingKMeans(_BaseKMeans): """Bisecting K-Means clustering. Read more in the :ref:`User Guide `. .. versionadded:: 1.1 Parameters ---------- n_clusters : int, default=8 The number of clusters to form as well as the number of centroids to generate. init : {'k-means++', 'random'} or callable, default='random' Method for initialization: 'k-means++' : selects initial cluster centers for k-mean clustering in a smart way to speed up convergence. See section Notes in k_init for more details. 'random': choose `n_clusters` observations (rows) at random from data for the initial centroids. If a callable is passed, it should take arguments X, n_clusters and a random state and return an initialization. n_init : int, default=1 Number of time the inner k-means algorithm will be run with different centroid seeds in each bisection. That will result producing for each bisection best output of n_init consecutive runs in terms of inertia. random_state : int, RandomState instance or None, default=None Determines random number generation for centroid initialization in inner K-Means. Use an int to make the randomness deterministic. See :term:`Glossary `. max_iter : int, default=300 Maximum number of iterations of the inner k-means algorithm at each bisection. verbose : int, default=0 Verbosity mode. tol : float, default=1e-4 Relative tolerance with regards to Frobenius norm of the difference in the cluster centers of two consecutive iterations to declare convergence. Used in inner k-means algorithm at each bisection to pick best possible clusters. copy_x : bool, default=True When pre-computing distances it is more numerically accurate to center the data first. If copy_x is True (default), then the original data is not modified. If False, the original data is modified, and put back before the function returns, but small numerical differences may be introduced by subtracting and then adding the data mean. Note that if the original data is not C-contiguous, a copy will be made even if copy_x is False. If the original data is sparse, but not in CSR format, a copy will be made even if copy_x is False. algorithm : {"lloyd", "elkan"}, default="lloyd" Inner K-means algorithm used in bisection. The classical EM-style algorithm is `"lloyd"`. The `"elkan"` variation can be more efficient on some datasets with well-defined clusters, by using the triangle inequality. However it's more memory intensive due to the allocation of an extra array of shape `(n_samples, n_clusters)`. bisecting_strategy : {"biggest_inertia", "largest_cluster"},\ default="biggest_inertia" Defines how bisection should be performed: - "biggest_inertia" means that BisectingKMeans will always check all calculated cluster for cluster with biggest SSE (Sum of squared errors) and bisect it. This approach concentrates on precision, but may be costly in terms of execution time (especially for larger amount of data points). - "largest_cluster" - BisectingKMeans will always split cluster with largest amount of points assigned to it from all clusters previously calculated. That should work faster than picking by SSE ('biggest_inertia') and may produce similar results in most cases. Attributes ---------- cluster_centers_ : ndarray of shape (n_clusters, n_features) Coordinates of cluster centers. If the algorithm stops before fully converging (see ``tol`` and ``max_iter``), these will not be consistent with ``labels_``. labels_ : ndarray of shape (n_samples,) Labels of each point. inertia_ : float Sum of squared distances of samples to their closest cluster center, weighted by the sample weights if provided. n_features_in_ : int Number of features seen during :term:`fit`. feature_names_in_ : ndarray of shape (`n_features_in_`,) Names of features seen during :term:`fit`. Defined only when `X` has feature names that are all strings. See Also -------- KMeans : Original implementation of K-Means algorithm. Notes ----- It might be inefficient when n_cluster is less than 3, due to unnecessary calculations for that case. Examples -------- >>> from sklearn.cluster import BisectingKMeans >>> import numpy as np >>> X = np.array([[1, 2], [1, 4], [1, 0], ... [10, 2], [10, 4], [10, 0], ... [10, 6], [10, 8], [10, 10]]) >>> bisect_means = BisectingKMeans(n_clusters=3, random_state=0).fit(X) >>> bisect_means.labels_ array([2, 2, 2, 0, 0, 0, 1, 1, 1], dtype=int32) >>> bisect_means.predict([[0, 0], [12, 3]]) array([2, 0], dtype=int32) >>> bisect_means.cluster_centers_ array([[10., 2.], [10., 8.], [ 1., 2.]]) """ _parameter_constraints: dict = { **_BaseKMeans._parameter_constraints, "init": [StrOptions({"k-means++", "random"}), callable], "copy_x": ["boolean"], "algorithm": [StrOptions({"lloyd", "elkan"})], "bisecting_strategy": [StrOptions({"biggest_inertia", "largest_cluster"})], } def __init__( self, n_clusters=8, *, init="random", n_init=1, random_state=None, max_iter=300, verbose=0, tol=1e-4, copy_x=True, algorithm="lloyd", bisecting_strategy="biggest_inertia", ): super().__init__( n_clusters=n_clusters, init=init, max_iter=max_iter, verbose=verbose, random_state=random_state, tol=tol, n_init=n_init, ) self.copy_x = copy_x self.algorithm = algorithm self.bisecting_strategy = bisecting_strategy def _warn_mkl_vcomp(self, n_active_threads): """Warn when vcomp and mkl are both present""" warnings.warn( "BisectingKMeans is known to have a memory leak on Windows " "with MKL, when there are less chunks than available " "threads. You can avoid it by setting the environment" f" variable OMP_NUM_THREADS={n_active_threads}." ) def _inertia_per_cluster(self, X, centers, labels, sample_weight): """Calculate the sum of squared errors (inertia) per cluster. Parameters ---------- X : {ndarray, csr_matrix} of shape (n_samples, n_features) The input samples. centers : ndarray of shape (n_clusters, n_features) The cluster centers. labels : ndarray of shape (n_samples,) Index of the cluster each sample belongs to. sample_weight : ndarray of shape (n_samples,) The weights for each observation in X. Returns ------- inertia_per_cluster : ndarray of shape (n_clusters,) Sum of squared errors (inertia) for each cluster. """ _inertia = _inertia_sparse if sp.issparse(X) else _inertia_dense inertia_per_cluster = np.empty(centers.shape[1]) for label in range(centers.shape[0]): inertia_per_cluster[label] = _inertia( X, sample_weight, centers, labels, self._n_threads, single_label=label ) return inertia_per_cluster def _bisect(self, X, x_squared_norms, sample_weight, cluster_to_bisect): """Split a cluster into 2 subsclusters. Parameters ---------- X : {ndarray, csr_matrix} of shape (n_samples, n_features) Training instances to cluster. x_squared_norms : ndarray of shape (n_samples,) Squared euclidean norm of each data point. sample_weight : ndarray of shape (n_samples,) The weights for each observation in X. cluster_to_bisect : _BisectingTree node object The cluster node to split. """ X = X[cluster_to_bisect.indices] x_squared_norms = x_squared_norms[cluster_to_bisect.indices] sample_weight = sample_weight[cluster_to_bisect.indices] best_inertia = None # Split samples in X into 2 clusters. # Repeating `n_init` times to obtain best clusters for _ in range(self.n_init): centers_init = self._init_centroids( X, x_squared_norms, self.init, self._random_state, n_centroids=2 ) labels, inertia, centers, _ = self._kmeans_single( X, sample_weight, centers_init, max_iter=self.max_iter, verbose=self.verbose, tol=self.tol, n_threads=self._n_threads, ) # allow small tolerance on the inertia to accommodate for # non-deterministic rounding errors due to parallel computation if best_inertia is None or inertia < best_inertia * (1 - 1e-6): best_labels = labels best_centers = centers best_inertia = inertia if self.verbose: print(f"New centroids from bisection: {best_centers}") if self.bisecting_strategy == "biggest_inertia": scores = self._inertia_per_cluster( X, best_centers, best_labels, sample_weight ) else: # bisecting_strategy == "largest_cluster" # Using minlength to make sure that we have the counts for both labels even # if all samples are labelled 0. scores = np.bincount(best_labels, minlength=2) cluster_to_bisect.split(best_labels, best_centers, scores) def fit(self, X, y=None, sample_weight=None): """Compute bisecting k-means clustering. Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) Training instances to cluster. .. note:: The data will be converted to C ordering, which will cause a memory copy if the given data is not C-contiguous. y : Ignored Not used, present here for API consistency by convention. sample_weight : array-like of shape (n_samples,), default=None The weights for each observation in X. If None, all observations are assigned equal weight. Returns ------- self Fitted estimator. """ self._validate_params() X = self._validate_data( X, accept_sparse="csr", dtype=[np.float64, np.float32], order="C", copy=self.copy_x, accept_large_sparse=False, ) self._check_params_vs_input(X) self._random_state = check_random_state(self.random_state) sample_weight = _check_sample_weight(sample_weight, X, dtype=X.dtype) self._n_threads = _openmp_effective_n_threads() if self.algorithm == "lloyd" or self.n_clusters == 1: self._kmeans_single = _kmeans_single_lloyd self._check_mkl_vcomp(X, X.shape[0]) else: self._kmeans_single = _kmeans_single_elkan # Subtract of mean of X for more accurate distance computations if not sp.issparse(X): self._X_mean = X.mean(axis=0) X -= self._X_mean # Initialize the hierarchical clusters tree self._bisecting_tree = _BisectingTree( indices=np.arange(X.shape[0]), center=X.mean(axis=0), score=0, ) x_squared_norms = row_norms(X, squared=True) for _ in range(self.n_clusters - 1): # Chose cluster to bisect cluster_to_bisect = self._bisecting_tree.get_cluster_to_bisect() # Split this cluster into 2 subclusters self._bisect(X, x_squared_norms, sample_weight, cluster_to_bisect) # Aggregate final labels and centers from the bisecting tree self.labels_ = np.full(X.shape[0], -1, dtype=np.int32) self.cluster_centers_ = np.empty((self.n_clusters, X.shape[1]), dtype=X.dtype) for i, cluster_node in enumerate(self._bisecting_tree.iter_leaves()): self.labels_[cluster_node.indices] = i self.cluster_centers_[i] = cluster_node.center cluster_node.label = i # label final clusters for future prediction cluster_node.indices = None # release memory # Restore original data if not sp.issparse(X): X += self._X_mean self.cluster_centers_ += self._X_mean _inertia = _inertia_sparse if sp.issparse(X) else _inertia_dense self.inertia_ = _inertia( X, sample_weight, self.cluster_centers_, self.labels_, self._n_threads ) self._n_features_out = self.cluster_centers_.shape[0] return self def predict(self, X): """Predict which cluster each sample in X belongs to. Prediction is made by going down the hierarchical tree in searching of closest leaf cluster. In the vector quantization literature, `cluster_centers_` is called the code book and each value returned by `predict` is the index of the closest code in the code book. Parameters ---------- X : {array-like, sparse matrix} of shape (n_samples, n_features) New data to predict. Returns ------- labels : ndarray of shape (n_samples,) Index of the cluster each sample belongs to. """ check_is_fitted(self) X = self._check_test_data(X) x_squared_norms = row_norms(X, squared=True) # sample weights are unused but necessary in cython helpers sample_weight = np.ones_like(x_squared_norms) labels = self._predict_recursive(X, sample_weight, self._bisecting_tree) return labels def _predict_recursive(self, X, sample_weight, cluster_node): """Predict recursively by going down the hierarchical tree. Parameters ---------- X : {ndarray, csr_matrix} of shape (n_samples, n_features) The data points, currently assigned to `cluster_node`, to predict between the subclusters of this node. sample_weight : ndarray of shape (n_samples,) The weights for each observation in X. cluster_node : _BisectingTree node object The cluster node of the hierarchical tree. Returns ------- labels : ndarray of shape (n_samples,) Index of the cluster each sample belongs to. """ if cluster_node.left is None: # This cluster has no subcluster. Labels are just the label of the cluster. return np.full(X.shape[0], cluster_node.label, dtype=np.int32) # Determine if data points belong to the left or right subcluster centers = np.vstack((cluster_node.left.center, cluster_node.right.center)) if hasattr(self, "_X_mean"): centers += self._X_mean cluster_labels = _labels_inertia_threadpool_limit( X, sample_weight, centers, self._n_threads, return_inertia=False, ) mask = cluster_labels == 0 # Compute the labels for each subset of the data points. labels = np.full(X.shape[0], -1, dtype=np.int32) labels[mask] = self._predict_recursive( X[mask], sample_weight[mask], cluster_node.left ) labels[~mask] = self._predict_recursive( X[~mask], sample_weight[~mask], cluster_node.right ) return labels def _more_tags(self): return {"preserves_dtype": [np.float64, np.float32]}