scikit-network 0.33.3__cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl

sknetwork/topology/core.pyx

@@ -0,0 +1,90 @@
+# distutils: language = c++
+# cython: language_level=3
+"""
+Created in June 2020
+@author: Julien Simonnet <julien.simonnet@etu.upmc.fr>
+@author: Yohann Robert <yohann.robert@etu.upmc.fr>
+"""
+cimport cython
+
+from typing import Union
+
+import numpy as np
+cimport numpy as np
+from scipy import sparse
+
+from sknetwork.utils.check import check_format
+from sknetwork.topology.minheap cimport MinHeap
+
+
+@cython.boundscheck(False)
+@cython.wraparound(False)
+cdef compute_core(int[:] indptr, int[:] indices):
+    """Compute the core value of each node.
+
+    Parameters
+    ----------
+    indptr :
+        CSR format index array of the adjacency matrix.
+    indices :
+        CSR format index pointer array of the adjacency matrix.
+
+    Returns
+    -------
+    labels :
+        Core value of each node.
+    """
+    cdef int n = indptr.shape[0] - 1
+    cdef int core_value	= 0	# current/max core value of the graph
+    cdef int min_node		# current node of minimum degree
+    cdef int i, j, k
+    cdef int[:] degrees = np.asarray(indptr)[1:] - np.asarray(indptr)[:n]
+    cdef np.ndarray[int, ndim=1] labels = np.empty((n,), dtype=np.int32)
+    cdef MinHeap mh = MinHeap.__new__(MinHeap, n)	# minimum heap with an update system
+
+    # insert all nodes in the heap
+    for i in range(n):
+        mh.insert_key(i, degrees)
+
+    i = n - 1
+    while not mh.empty():
+        min_node = mh.pop_min(degrees)
+        core_value = max(core_value, degrees[min_node])
+
+        # decrease the degree of each neighbor of min_node
+        for k in range(indptr[min_node], indptr[min_node+1]):
+            j = indices[k]
+            degrees[j] -= 1
+            mh.decrease_key(j, degrees)		# update the heap to take the new degree into account
+
+        labels[min_node] = core_value
+        i -= 1
+
+    return np.asarray(labels)
+
+
+def get_core_decomposition(adjacency: Union[np.ndarray, sparse.csr_matrix]) -> np.ndarray:
+    """Get the k-core decomposition of a graph.
+
+    Parameters
+    ----------
+    adjacency :
+        Adjacency matrix of the graph.
+
+    Returns
+    -------
+    core_values :
+         Core value of each node.
+
+    Example
+    -------
+    >>> from sknetwork.data import karate_club
+    >>> adjacency = karate_club()
+    >>> core_values = get_core_decomposition(adjacency)
+    >>> len(core_values)
+    34
+    """
+    adjacency = check_format(adjacency, allow_empty=True)
+    indptr = adjacency.indptr
+    indices = adjacency.indices
+    return compute_core(indptr, indices)

sknetwork/topology/cycles.py

@@ -0,0 +1,243 @@
+#!/usr/bin/env python3
+# -*- coding: utf-8 -*-
+"""
+Created in February 2024
+@author: Thomas Bonald <thomas.bonald@telecom-paris.fr>
+@author: Yiwen Peng <yiwen.peng@telecom-paris.fr>
+"""
+from typing import Tuple, Optional, Union, List
+
+import numpy as np
+from scipy import sparse
+
+from sknetwork.utils.check import is_symmetric, check_format
+from sknetwork.utils.format import get_adjacency
+from sknetwork.path import get_distances
+
+
+def is_acyclic(adjacency: sparse.csr_matrix, directed: Optional[bool] = None) -> bool:
+    """Check whether a graph has no cycle.
+
+    Parameters
+    ----------
+    adjacency:
+        Adjacency matrix of the graph.
+    directed:
+        Whether to consider the graph as directed (inferred if not specified).
+    Returns
+    -------
+    is_acyclic : bool
+        A boolean with value True if the graph has no cycle and False otherwise.
+
+    Example
+    -------
+    >>> from sknetwork.topology import is_acyclic
+    >>> from sknetwork.data import star, grid
+    >>> is_acyclic(star())
+    True
+    >>> is_acyclic(grid())
+    False
+    """
+    if directed is False:
+        # the graph must be undirected
+        if not is_symmetric(adjacency):
+            raise ValueError("The adjacency matrix is not symmetric. The parameter 'directed' must be True.")
+    elif directed is None:
+        # if not specified, infer from the graph
+        directed = not is_symmetric(adjacency)
+    has_loops = (adjacency.diagonal() > 0).any()
+    if has_loops:
+        return False
+    else:
+        n_cc = sparse.csgraph.connected_components(adjacency, directed, connection='strong', return_labels=False)
+        n_nodes = adjacency.shape[0]
+        if directed:
+            return n_cc == n_nodes
+        else:
+            n_edges = adjacency.nnz // 2
+            return n_cc == n_nodes - n_edges
+
+
+def get_cycles(adjacency: sparse.csr_matrix, directed: Optional[bool] = None) -> List[List[int]]:
+    """Get all possible cycles of a graph.
+
+    Parameters
+    ----------
+    adjacency :
+        Adjacency matrix of the graph.
+    directed :
+        Whether to consider the graph as directed (inferred if not specified).
+
+    Returns
+    -------
+    cycles : list
+        List of cycles, each cycle represented as a list of nodes.
+
+    Example
+    -------
+    >>> from sknetwork.topology import get_cycles
+    >>> from sknetwork.data import cyclic_digraph
+    >>> graph = cyclic_digraph(4, metadata=True)
+    >>> len(get_cycles(graph.adjacency, directed=True))
+    1
+    """
+
+    if directed is False:
+        # the graph must be undirected
+        if not is_symmetric(adjacency):
+            raise ValueError("The adjacency matrix is not symmetric. The parameter 'directed' must be True.")
+    elif directed is None:
+        # if not specified, infer from the graph
+        directed = not is_symmetric(adjacency)
+
+    cycles = []
+    n_nodes = adjacency.shape[0]
+    self_loops = np.argwhere(adjacency.diagonal() > 0).ravel()
+    if len(self_loops) > 0:
+        # add self-loops as cycles
+        for node in self_loops:
+            cycles.append([node])
+
+    # check if the graph is acyclic
+    n_cc, cc_labels = sparse.csgraph.connected_components(adjacency, directed, connection='strong', return_labels=True)
+    if directed and n_cc == n_nodes:
+        return cycles
+    elif not directed:
+        # acyclic undirected graph
+        n_edges = adjacency.nnz // 2
+        if n_cc == n_nodes - n_edges:
+            return cycles
+
+    # locate possible cycles
+    labels, counts = np.unique(cc_labels, return_counts=True)
+    if directed:
+        labels = labels[counts > 1]
+    cycle_starts = [np.argwhere(cc_labels == label).ravel()[0] for label in labels]
+
+    # find cycles for indicated nodes (depth-first traversal)
+    for start_node in cycle_starts:
+        stack = [(start_node, [start_node])]
+        while stack:
+            current_node, path = stack.pop()
+            for neighbor in adjacency.indices[adjacency.indptr[current_node]:adjacency.indptr[current_node+1]]:
+                if not directed and len(path) > 1 and neighbor == path[-2]:
+                    # ignore the inverse edge (back move) in undirected graph
+                    continue
+                if neighbor in path:
+                    cycles.append(path[path.index(neighbor):])
+                else:
+                    stack.append((neighbor, path + [neighbor]))
+
+    # remove duplicates
+    visited_cycles, unique_cycles = set(), []
+    for cycle in cycles:
+        candidate = np.roll(cycle, -cycle.index(min(cycle)))
+        current_cycle = tuple(candidate)
+        if not directed:
+            current_cycle = tuple(np.sort(candidate))
+        if current_cycle not in visited_cycles:
+            unique_cycles.append(list(candidate))
+        visited_cycles.add(current_cycle)
+
+    return unique_cycles
+
+
+def break_cycles(adjacency: sparse.csr_matrix, root: Union[int, List[int]],
+                 directed: Optional[bool] = None) -> sparse.csr_matrix:
+    """Break cycles of a graph from given roots.
+
+    Parameters
+    ----------
+    adjacency :
+        Adjacency matrix of the graph.
+    root :
+        The root node or list of root nodes to break cycles from.
+    directed :
+        Whether to consider the graph as directed (inferred if not specified).
+
+    Returns
+    -------
+    adjacency : sparse.csr_matrix
+        Adjacency matrix of the acyclic graph.
+
+    Example
+    -------
+    >>> from sknetwork.topology import break_cycles, is_acyclic
+    >>> from sknetwork.data import cyclic_digraph
+    >>> adjacency = cyclic_digraph(4)
+    >>> dag = break_cycles(adjacency, root=0, directed=True)
+    >>> is_acyclic(dag, directed=True)
+    True
+    """
+
+    if is_acyclic(adjacency, directed):
+        return adjacency
+
+    if root is None:
+        raise ValueError("The parameter root must be specified.")
+    else:
+        out_degree = len(adjacency[root].indices)
+        if out_degree == 0:
+            raise ValueError("Invalid root node. The root must have at least one outgoing edge.")
+        if isinstance(root, int):
+            root = [root]
+
+    if directed is False:
+        # the graph must be undirected
+        if not is_symmetric(adjacency):
+            raise ValueError("The adjacency matrix is not symmetric. The parameter 'directed' must be True.")
+    elif directed is None:
+        # if not specified, infer from the graph
+        directed = not is_symmetric(adjacency)
+
+    # break self-loops
+    adjacency = (adjacency - sparse.diags(adjacency.diagonal().astype(int), format='csr')).astype(int)
+
+    if directed:
+        # break cycles from the cycle node closest to the root
+        _, cc_labels = sparse.csgraph.connected_components(adjacency, directed, connection='strong', return_labels=True)
+        labels, counts = np.unique(cc_labels, return_counts=True)
+        cycle_labels = labels[counts > 1]
+        distances = get_distances(adjacency, source=root)
+
+        for label in cycle_labels:
+            cycle_nodes = np.argwhere(cc_labels == label).ravel()
+            roots_ix = np.argwhere(distances[cycle_nodes] == min(distances[cycle_nodes])).ravel()
+            subroots = set(cycle_nodes[roots_ix])
+            stack = [(subroot, [subroot]) for subroot in subroots]
+            # break cycles using depth-first traversal
+            while stack:
+                current_node, path = stack.pop()
+                # check if the edge still exists
+                if len(path) > 1 and adjacency[path[-2], path[-1]] <= 0:
+                    continue
+                neighbors = adjacency.indices[adjacency.indptr[current_node]:adjacency.indptr[current_node+1]]
+                cycle_neighbors = set(neighbors) & set(cycle_nodes)
+                for neighbor in cycle_neighbors:
+                    if neighbor in path:
+                        adjacency[current_node, neighbor] = 0
+                        adjacency.eliminate_zeros()
+                    else:
+                        stack.append((neighbor, path + [neighbor]))
+    else:
+        # break cycles from given roots for undirected graphs
+        for start_node in root:
+            stack = [(start_node, [start_node])]
+            # break cycles using depth-first traversal
+            while stack:
+                current_node, path = stack.pop()
+                # check if the edge still exists
+                if len(path) > 1 and adjacency[path[-2], path[-1]] <= 0:
+                    continue
+                neighbors = list(adjacency.indices[adjacency.indptr[current_node]:adjacency.indptr[current_node+1]])
+                for neighbor in neighbors:
+                    if len(path) > 1 and neighbor == path[-2]:
+                        continue
+                    if neighbor in path:
+                        adjacency[current_node, neighbor] = 0
+                        adjacency[neighbor, current_node] = 0
+                        adjacency.eliminate_zeros()
+                    else:
+                        stack.append((neighbor, path + [neighbor]))
+
+    return adjacency