Source code for sknetwork.hierarchy.postprocess

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on June 2019
@author: Thomas Bonald <bonald@enst.fr>
@author: Bertrand Charpentier <bertrand.charpentier@live.fr>
@author: Quentin Lutz <qlutz@enst.fr>
"""

import copy
from collections import defaultdict
from typing import Optional, Union, Tuple

import numpy as np

from sknetwork.utils.check import check_n_clusters, check_dendrogram


[docs] def reorder_dendrogram(dendrogram: np.ndarray) -> np.ndarray: """Reorder the dendrogram in non-decreasing order of height.""" n = dendrogram.shape[0] + 1 order = np.zeros((2, n - 1), float) order[0] = np.max(dendrogram[:, :2], axis=1) order[1] = dendrogram[:, 2] index = np.lexsort(order) dendrogram_new = dendrogram[index] index_new = np.arange(2 * n - 1) index_new[n + index] = np.arange(n, 2 * n - 1) dendrogram_new[:, 0] = index_new[dendrogram_new[:, 0].astype(int)] dendrogram_new[:, 1] = index_new[dendrogram_new[:, 1].astype(int)] return dendrogram_new
def get_labels(dendrogram: np.ndarray, cluster: dict, sort_clusters: bool, return_dendrogram: bool): """Returns the labels from clusters.""" n = len(dendrogram) + 1 clusters = list(cluster.values()) if sort_clusters: sizes = np.array([len(nodes) for nodes in clusters]) index = np.argsort(-sizes) clusters = [clusters[i] for i in index] labels = np.zeros(n, dtype=int) for label, nodes in enumerate(clusters): labels[nodes] = label if return_dendrogram: cluster_index = {i: label for i, label in enumerate(labels)} cluster_size = {i: len(cluster) for i, cluster in enumerate(clusters)} dendrogram_new = [] current_cluster = len(labels) current_cluster_new = len(clusters) for i, j, height, _ in dendrogram: i_new = cluster_index.pop(int(i)) j_new = cluster_index.pop(int(j)) if i_new != j_new: size = cluster_size.pop(i_new) + cluster_size.pop(j_new) cluster_size[current_cluster_new] = size cluster_index[current_cluster] = current_cluster_new dendrogram_new.append([i_new, j_new, height, size]) current_cluster_new += 1 else: cluster_index[current_cluster] = i_new current_cluster += 1 dendrogram_new = np.array(dendrogram_new) return labels, dendrogram_new else: return labels
[docs] def cut_straight(dendrogram: np.ndarray, n_clusters: Optional[int] = None, threshold: Optional[float] = None, sort_clusters: bool = True, return_dendrogram: bool = False) \ -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]: """Cut a dendrogram and return the corresponding clustering. Parameters ---------- dendrogram : np.ndarray Dendrogram. n_clusters : int Number of clusters (optional). The number of clusters can be larger than n_clusters in case of equal heights in the dendrogram. threshold : float Threshold on height (optional). If both n_clusters and threshold are ``None``, n_clusters is set to 2. sort_clusters : bool If ``True``, sorts clusters in decreasing order of size. return_dendrogram : bool If ``True``, returns the dendrogram formed by the clusters up to the root. Returns ------- labels : np.ndarray Cluster of each node. dendrogram_aggregate : np.ndarray Dendrogram starting from clusters (leaves = clusters). Example ------- >>> from sknetwork.hierarchy import cut_straight >>> dendrogram = np.array([[0, 1, 0, 2], [2, 3, 1, 3]]) >>> cut_straight(dendrogram) array([0, 0, 1]) """ check_dendrogram(dendrogram) n = dendrogram.shape[0] + 1 if return_dendrogram: height = dendrogram[:, 2] if not np.all(height[:-1] <= height[1:]): dendrogram = reorder_dendrogram(dendrogram) cluster = {i: [i] for i in range(n)} if n_clusters is None: if threshold is None: n_clusters = 2 else: n_clusters = n else: check_n_clusters(n_clusters, n, n_min=1) cut = np.sort(dendrogram[:, 2])[n - n_clusters] if threshold is not None: cut = max(cut, threshold) for t in range(n - 1): i = int(dendrogram[t][0]) j = int(dendrogram[t][1]) if dendrogram[t][2] < cut and i in cluster and j in cluster: cluster[n + t] = cluster.pop(i) + cluster.pop(j) return get_labels(dendrogram, cluster, sort_clusters, return_dendrogram)
[docs] def cut_balanced(dendrogram: np.ndarray, max_cluster_size: int = 20, sort_clusters: bool = True, return_dendrogram: bool = False) -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]: """Cuts a dendrogram with a constraint on the cluster size and returns the corresponding clustering. Parameters ---------- dendrogram : np.ndarray Dendrogram max_cluster_size : int Maximum size of each cluster. sort_clusters : bool If ``True``, sort labels in decreasing order of cluster size. return_dendrogram : bool If ``True``, returns the dendrogram formed by the clusters up to the root. Returns ------- labels : np.ndarray Label of each node. dendrogram_aggregate : np.ndarray Dendrogram starting from clusters (leaves = clusters). Example ------- >>> from sknetwork.hierarchy import cut_balanced >>> dendrogram = np.array([[0, 1, 0, 2], [2, 3, 1, 3]]) >>> cut_balanced(dendrogram, 2) array([0, 0, 1]) """ check_dendrogram(dendrogram) n = dendrogram.shape[0] + 1 if max_cluster_size < 2 or max_cluster_size > n: raise ValueError("The maximum cluster size must be between 2 and the number of nodes.") cluster = {i: [i] for i in range(n)} for t in range(n - 1): i = int(dendrogram[t][0]) j = int(dendrogram[t][1]) if i in cluster and j in cluster and len(cluster[i]) + len(cluster[j]) <= max_cluster_size: cluster[n + t] = cluster.pop(i) + cluster.pop(j) return get_labels(dendrogram, cluster, sort_clusters, return_dendrogram)
[docs] def aggregate_dendrogram(dendrogram: np.ndarray, n_clusters: int = 2, return_counts: bool = False) \ -> Union[np.ndarray, Tuple[np.ndarray, np.ndarray]]: """Aggregate a dendrogram in order to get a certain number of leaves. The leaves in the output dendrogram correspond to subtrees in the input one. Parameters ---------- dendrogram : np.ndarray The input to aggregate. n_clusters : int Number of clusters (or leaves) to keep. return_counts : bool If ``True``, returns an array of counts corresponding to the sizes of the merged subtrees. The sum of the counts is equal to the number of samples in the input dendrogram. Returns ------- new_dendrogram : np.ndarray Aggregated dendrogram. The nodes are reindexed from 0. counts : np.ndarray Size of the subtrees corresponding to each leaf in new_dendrogram. """ n_nodes: int = dendrogram.shape[0] + 1 check_n_clusters(n_clusters, n_nodes, n_min=1) new_dendrogram = dendrogram[n_nodes - n_clusters:].copy() node_indices = np.array(sorted(set(new_dendrogram[:, 0]).union(set(new_dendrogram[:, 1])))) new_index = {ix: i for i, ix in enumerate(node_indices)} for j in range(2): for i in range(new_dendrogram.shape[0]): new_dendrogram[i, j] = new_index[new_dendrogram[i, j]] if return_counts: leaves = node_indices[:n_clusters].astype(int) leaves_indices = leaves - n_nodes counts = dendrogram[leaves_indices, 3] return new_dendrogram, counts.astype(int) else: return new_dendrogram
def get_index(tree): """Reindex a dendrogram from the leaves Parameters ---------- tree : The tree to be indexed Returns ------- index : The index of the root of the given tree """ if type(tree) != list: return tree else: return np.max([get_index(t) for t in tree]) def get_dendrogram(tree, dendrogram=None, index=None, depth=0, size=None, copy_tree=False): """Get dendrogram from tree. Parameters ---------- tree : The initial tree dendrogram : Intermediary dendrogram for recursive use index : Intermediary index for recursive use depth : Current depth for recursive use size : Current leaf count for recursive use copy_tree : If ``True``, ensure the passed tree remains unchanged. Returns ------- dendrogram`: The reordered dendrogram index : The indexing array """ if copy_tree: return get_dendrogram(copy.deepcopy(tree)) else: if dendrogram is None: dendrogram = [] if index is None: index = get_index(tree) if size is None: size = defaultdict(lambda: 1) if len(tree) > 1: lengths = np.array([len(t) for t in tree]) if np.max(lengths) == 1: # merge all i = tree.pop()[0] j = tree.pop()[0] s = size[i] + size[j] dendrogram.append([i, j, float(-depth), s]) index += 1 while len(tree): s += 1 dendrogram.append([index, tree.pop()[0], float(-depth), s]) index += 1 size[index] = s tree.append(index) return dendrogram, index else: i = np.argwhere(lengths > 1).ravel()[0] dendrogram_, index_ = get_dendrogram(tree[i], None, index, depth + 1, size) dendrogram += dendrogram_ return get_dendrogram(tree, dendrogram, index_, depth, size) else: return dendrogram, index def split_dendrogram(dendrogram: np.ndarray, shape: tuple): """Split the dendrogram of a bipartite graph into 2 dendrograms, one for each part. Parameters ---------- dendrogram : Dendrogram of the bipartite graph. shape : Shape of the biadjacency matrix. Returns ------- dendrogram_row : Dendrogram for the rows. dendrogram_col : Dendrogram for the columns. """ n1, n2 = shape dendrogram_row = [] dendrogram_col = [] id_row_new = n1 id_col_new = n2 size_row = {i: 1 for i in range(n1)} size_col = {i + n1: 1 for i in range(n2)} id_row = {i: i for i in range(n1)} id_col = {i + n1: i for i in range(n2)} for t in range(n1 + n2 - 1): i = dendrogram[t, 0] j = dendrogram[t, 1] if i in id_row and j in id_row: size_row[n1 + n2 + t] = size_row.pop(i) + size_row.pop(j) id_row[n1 + n2 + t] = id_row_new dendrogram_row.append([id_row.pop(i), id_row.pop(j), dendrogram[t, 2], size_row[n1 + n2 + t]]) id_row_new += 1 elif i in id_row: size_row[n1 + n2 + t] = size_row.pop(i) id_row[n1 + n2 + t] = id_row.pop(i) elif j in id_row: size_row[n1 + n2 + t] = size_row.pop(j) id_row[n1 + n2 + t] = id_row.pop(j) if i in id_col and j in id_col: size_col[n1 + n2 + t] = size_col.pop(i) + size_col.pop(j) id_col[n1 + n2 + t] = id_col_new dendrogram_col.append([id_col.pop(i), id_col.pop(j), dendrogram[t, 2], size_col[n1 + n2 + t]]) id_col_new += 1 elif i in id_col: size_col[n1 + n2 + t] = size_col.pop(i) id_col[n1 + n2 + t] = id_col.pop(i) elif j in id_col: size_col[n1 + n2 + t] = size_col.pop(j) id_col[n1 + n2 + t] = id_col.pop(j) return np.array(dendrogram_row), np.array(dendrogram_col)