# Authors: Manoj Kumar <manojkumarsivaraj334@gmail.com>
# Alexandre Gramfort <alexandre.gramfort@telecom-paristech.fr>
# Joel Nothman <joel.nothman@gmail.com>
# License: BSD 3 clause
import warnings
import numbers
import numpy as np
from scipy import sparse
from math import sqrt
from ..metrics import pairwise_distances_argmin
from ..metrics.pairwise import euclidean_distances
from ..base import TransformerMixin, ClusterMixin, BaseEstimator
from ..utils import check_array
from ..utils.extmath import row_norms
from ..utils.validation import check_is_fitted, _deprecate_positional_args
from ..exceptions import ConvergenceWarning
from . import AgglomerativeClustering
def _iterate_sparse_X(X):
"""This little hack returns a densified row when iterating over a sparse
matrix, instead of constructing a sparse matrix for every row that is
expensive.
"""
n_samples = X.shape[0]
X_indices = X.indices
X_data = X.data
X_indptr = X.indptr
for i in range(n_samples):
row = np.zeros(X.shape[1])
startptr, endptr = X_indptr[i], X_indptr[i + 1]
nonzero_indices = X_indices[startptr:endptr]
row[nonzero_indices] = X_data[startptr:endptr]
yield row
def _split_node(node, threshold, branching_factor):
"""The node has to be split if there is no place for a new subcluster
in the node.
1. Two empty nodes and two empty subclusters are initialized.
2. The pair of distant subclusters are found.
3. The properties of the empty subclusters and nodes are updated
according to the nearest distance between the subclusters to the
pair of distant subclusters.
4. The two nodes are set as children to the two subclusters.
"""
new_subcluster1 = _CFSubcluster()
new_subcluster2 = _CFSubcluster()
new_node1 = _CFNode(
threshold=threshold, branching_factor=branching_factor,
is_leaf=node.is_leaf,
n_features=node.n_features)
new_node2 = _CFNode(
threshold=threshold, branching_factor=branching_factor,
is_leaf=node.is_leaf,
n_features=node.n_features)
new_subcluster1.child_ = new_node1
new_subcluster2.child_ = new_node2
if node.is_leaf:
if node.prev_leaf_ is not None:
node.prev_leaf_.next_leaf_ = new_node1
new_node1.prev_leaf_ = node.prev_leaf_
new_node1.next_leaf_ = new_node2
new_node2.prev_leaf_ = new_node1
new_node2.next_leaf_ = node.next_leaf_
if node.next_leaf_ is not None:
node.next_leaf_.prev_leaf_ = new_node2
dist = euclidean_distances(
node.centroids_, Y_norm_squared=node.squared_norm_, squared=True)
n_clusters = dist.shape[0]
farthest_idx = np.unravel_index(
dist.argmax(), (n_clusters, n_clusters))
node1_dist, node2_dist = dist[(farthest_idx,)]
node1_closer = node1_dist < node2_dist
for idx, subcluster in enumerate(node.subclusters_):
if node1_closer[idx]:
new_node1.append_subcluster(subcluster)
new_subcluster1.update(subcluster)
else:
new_node2.append_subcluster(subcluster)
new_subcluster2.update(subcluster)
return new_subcluster1, new_subcluster2
class _CFNode:
"""Each node in a CFTree is called a CFNode.
The CFNode can have a maximum of branching_factor
number of CFSubclusters.
Parameters
----------
threshold : float
Threshold needed for a new subcluster to enter a CFSubcluster.
branching_factor : int
Maximum number of CF subclusters in each node.
is_leaf : bool
We need to know if the CFNode is a leaf or not, in order to
retrieve the final subclusters.
n_features : int
The number of features.
Attributes
----------
subclusters_ : list
List of subclusters for a particular CFNode.
prev_leaf_ : _CFNode
Useful only if is_leaf is True.
next_leaf_ : _CFNode
next_leaf. Useful only if is_leaf is True.
the final subclusters.
init_centroids_ : ndarray of shape (branching_factor + 1, n_features)
Manipulate ``init_centroids_`` throughout rather than centroids_ since
the centroids are just a view of the ``init_centroids_`` .
init_sq_norm_ : ndarray of shape (branching_factor + 1,)
manipulate init_sq_norm_ throughout. similar to ``init_centroids_``.
centroids_ : ndarray of shape (branching_factor + 1, n_features)
View of ``init_centroids_``.
squared_norm_ : ndarray of shape (branching_factor + 1,)
View of ``init_sq_norm_``.
"""
def __init__(self, *, threshold, branching_factor, is_leaf, n_features):
self.threshold = threshold
self.branching_factor = branching_factor
self.is_leaf = is_leaf
self.n_features = n_features
# The list of subclusters, centroids and squared norms
# to manipulate throughout.
self.subclusters_ = []
self.init_centroids_ = np.zeros((branching_factor + 1, n_features))
self.init_sq_norm_ = np.zeros((branching_factor + 1))
self.squared_norm_ = []
self.prev_leaf_ = None
self.next_leaf_ = None
def append_subcluster(self, subcluster):
n_samples = len(self.subclusters_)
self.subclusters_.append(subcluster)
self.init_centroids_[n_samples] = subcluster.centroid_
self.init_sq_norm_[n_samples] = subcluster.sq_norm_
# Keep centroids and squared norm as views. In this way
# if we change init_centroids and init_sq_norm_, it is
# sufficient,
self.centroids_ = self.init_centroids_[:n_samples + 1, :]
self.squared_norm_ = self.init_sq_norm_[:n_samples + 1]
def update_split_subclusters(self, subcluster,
new_subcluster1, new_subcluster2):
"""Remove a subcluster from a node and update it with the
split subclusters.
"""
ind = self.subclusters_.index(subcluster)
self.subclusters_[ind] = new_subcluster1
self.init_centroids_[ind] = new_subcluster1.centroid_
self.init_sq_norm_[ind] = new_subcluster1.sq_norm_
self.append_subcluster(new_subcluster2)
def insert_cf_subcluster(self, subcluster):
"""Insert a new subcluster into the node."""
if not self.subclusters_:
self.append_subcluster(subcluster)
return False
threshold = self.threshold
branching_factor = self.branching_factor
# We need to find the closest subcluster among all the
# subclusters so that we can insert our new subcluster.
dist_matrix = np.dot(self.centroids_, subcluster.centroid_)
dist_matrix *= -2.
dist_matrix += self.squared_norm_
closest_index = np.argmin(dist_matrix)
closest_subcluster = self.subclusters_[closest_index]
# If the subcluster has a child, we need a recursive strategy.
if closest_subcluster.child_ is not None:
split_child = closest_subcluster.child_.insert_cf_subcluster(
subcluster)
if not split_child:
# If it is determined that the child need not be split, we
# can just update the closest_subcluster
closest_subcluster.update(subcluster)
self.init_centroids_[closest_index] = \
self.subclusters_[closest_index].centroid_
self.init_sq_norm_[closest_index] = \
self.subclusters_[closest_index].sq_norm_
return False
# things not too good. we need to redistribute the subclusters in
# our child node, and add a new subcluster in the parent
# subcluster to accommodate the new child.
else:
new_subcluster1, new_subcluster2 = _split_node(
closest_subcluster.child_, threshold, branching_factor)
self.update_split_subclusters(
closest_subcluster, new_subcluster1, new_subcluster2)
if len(self.subclusters_) > self.branching_factor:
return True
return False
# good to go!
else:
merged = closest_subcluster.merge_subcluster(
subcluster, self.threshold)
if merged:
self.init_centroids_[closest_index] = \
closest_subcluster.centroid_
self.init_sq_norm_[closest_index] = \
closest_subcluster.sq_norm_
return False
# not close to any other subclusters, and we still
# have space, so add.
elif len(self.subclusters_) < self.branching_factor:
self.append_subcluster(subcluster)
return False
# We do not have enough space nor is it closer to an
# other subcluster. We need to split.
else:
self.append_subcluster(subcluster)
return True
class _CFSubcluster:
"""Each subcluster in a CFNode is called a CFSubcluster.
A CFSubcluster can have a CFNode has its child.
Parameters
----------
linear_sum : ndarray of shape (n_features,), default=None
Sample. This is kept optional to allow initialization of empty
subclusters.
Attributes
----------
n_samples_ : int
Number of samples that belong to each subcluster.
linear_sum_ : ndarray
Linear sum of all the samples in a subcluster. Prevents holding
all sample data in memory.
squared_sum_ : float
Sum of the squared l2 norms of all samples belonging to a subcluster.
centroid_ : ndarray of shape (branching_factor + 1, n_features)
Centroid of the subcluster. Prevent recomputing of centroids when
``CFNode.centroids_`` is called.
child_ : _CFNode
Child Node of the subcluster. Once a given _CFNode is set as the child
of the _CFNode, it is set to ``self.child_``.
sq_norm_ : ndarray of shape (branching_factor + 1,)
Squared norm of the subcluster. Used to prevent recomputing when
pairwise minimum distances are computed.
"""
def __init__(self, *, linear_sum=None):
if linear_sum is None:
self.n_samples_ = 0
self.squared_sum_ = 0.0
self.centroid_ = self.linear_sum_ = 0
else:
self.n_samples_ = 1
self.centroid_ = self.linear_sum_ = linear_sum
self.squared_sum_ = self.sq_norm_ = np.dot(
self.linear_sum_, self.linear_sum_)
self.child_ = None
def update(self, subcluster):
self.n_samples_ += subcluster.n_samples_
self.linear_sum_ += subcluster.linear_sum_
self.squared_sum_ += subcluster.squared_sum_
self.centroid_ = self.linear_sum_ / self.n_samples_
self.sq_norm_ = np.dot(self.centroid_, self.centroid_)
def merge_subcluster(self, nominee_cluster, threshold):
"""Check if a cluster is worthy enough to be merged. If
yes then merge.
"""
new_ss = self.squared_sum_ + nominee_cluster.squared_sum_
new_ls = self.linear_sum_ + nominee_cluster.linear_sum_
new_n = self.n_samples_ + nominee_cluster.n_samples_
new_centroid = (1 / new_n) * new_ls
new_norm = np.dot(new_centroid, new_centroid)
dot_product = (-2 * new_n) * new_norm
sq_radius = (new_ss + dot_product) / new_n + new_norm
if sq_radius <= threshold ** 2:
(self.n_samples_, self.linear_sum_, self.squared_sum_,
self.centroid_, self.sq_norm_) = \
new_n, new_ls, new_ss, new_centroid, new_norm
return True
return False
@property
def radius(self):
"""Return radius of the subcluster"""
dot_product = -2 * np.dot(self.linear_sum_, self.centroid_)
return sqrt(
((self.squared_sum_ + dot_product) / self.n_samples_) +
self.sq_norm_)
class Birch(ClusterMixin, TransformerMixin, BaseEstimator):
"""Implements the Birch clustering algorithm.
It is a memory-efficient, online-learning algorithm provided as an
alternative to :class:`MiniBatchKMeans`. It constructs a tree
data structure with the cluster centroids being read off the leaf.
These can be either the final cluster centroids or can be provided as input
to another clustering algorithm such as :class:`AgglomerativeClustering`.
Read more in the :ref:`User Guide <birch>`.
.. versionadded:: 0.16
Parameters
----------
threshold : float, default=0.5
The radius of the subcluster obtained by merging a new sample and the
closest subcluster should be lesser than the threshold. Otherwise a new
subcluster is started. Setting this value to be very low promotes
splitting and vice-versa.
Loading ...