Add function to find densest subgraph

Fixes Qiskit#570

src/dense_subgraph.rs

@@ -0,0 +1,199 @@
+// Licensed under the Apache License, Version 2.0 (the "License"); you may
+// not use this file except in compliance with the License. You may obtain
+// a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
+// WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
+// License for the specific language governing permissions and limitations
+// under the License.
+
+use hashbrown::{HashMap, HashSet};
+
+use petgraph::graph::NodeIndex;
+use petgraph::prelude::*;
+use petgraph::visit::IntoEdgeReferences;
+use petgraph::EdgeType;
+
+use rayon::prelude::*;
+
+use numpy::ToPyArray;
+use pyo3::prelude::*;
+use pyo3::Python;
+
+use crate::digraph;
+use crate::graph;
+use crate::StablePyGraph;
+
+struct SubsetResult {
+    pub count: usize,
+    pub error: f64,
+    pub map: Vec<NodeIndex>,
+    pub subgraph: Vec<[NodeIndex; 2]>,
+}
+
+pub fn densest_subgraph<Ty>(
+    py: Python,
+    graph: &StablePyGraph<Ty>,
+    num_nodes: usize,
+    weight_callback: Option<PyObject>,
+) -> PyResult<(PyObject, PyObject, PyObject)>
+where
+    Ty: EdgeType + Sync,
+{
+    let node_indices: Vec<NodeIndex> = graph.node_indices().collect();
+    let float_callback =
+        |callback: PyObject, source_node: usize, target_node: usize| -> PyResult<f64> {
+            let res = callback.as_ref(py).call1((source_node, target_node))?;
+            res.extract()
+        };
+    let mut weight_map: Option<HashMap<[NodeIndex; 2], f64>> = None;
+
+    if weight_callback.is_some() {
+        let mut inner_weight_map: HashMap<[NodeIndex; 2], f64> =
+            HashMap::with_capacity(graph.edge_count());
+        let callback = weight_callback.as_ref().unwrap();
+        for edge in graph.edge_references() {
+            let source: NodeIndex = edge.source();
+            let target: NodeIndex = edge.target();
+            let weight = float_callback(callback.clone_ref(py), source.index(), target.index())?;
+            inner_weight_map.insert([source, target], weight);
+        }
+        weight_map = Some(inner_weight_map);
+    }
+    let reduce_identity_fn = || -> SubsetResult {
+        SubsetResult {
+            count: 0,
+            map: Vec::new(),
+            error: std::f64::INFINITY,
+            subgraph: Vec::new(),
+        }
+    };
+
+    let reduce_fn = |best: SubsetResult, curr: SubsetResult| -> SubsetResult {
+        if weight_callback.is_some() {
+            if curr.count >= best.count && curr.error <= best.error {
+                curr
+            } else {
+                best
+            }
+        } else if curr.count > best.count {
+            curr
+        } else {
+            best
+        }
+    };
+
+    let best_result = node_indices
+        .into_par_iter()
+        .map(|index| {
+            let mut subgraph: Vec<[NodeIndex; 2]> = Vec::with_capacity(num_nodes);
+            let mut bfs = Bfs::new(&graph, index);
+            let mut bfs_vec: Vec<NodeIndex> = Vec::with_capacity(num_nodes);
+            let mut bfs_set: HashSet<NodeIndex> = HashSet::with_capacity(num_nodes);
+
+            let mut count = 0;
+            while let Some(node) = bfs.next(&graph) {
+                bfs_vec.push(node);
+                bfs_set.insert(node);
+                count += 1;
+                if count >= num_nodes {
+                    break;
+                }
+            }
+            let mut connection_count = 0;
+            for node in &bfs_vec {
+                for j in graph.node_indices() {
+                    if graph.contains_edge(*node, j) && bfs_set.contains(&j){
+                        connection_count += 1;
+                        subgraph.push([*node, j]);
+                    }
+                }
+            }
+            let error = match &weight_map {
+                Some(map) => subgraph.iter().map(|edge| map[edge]).sum::<f64>() / num_nodes as f64,
+                None => 0.,
+            };
+            SubsetResult {
+                count: connection_count,
+                error,
+                map: bfs_vec,
+                subgraph,
+            }
+        })
+        .reduce(reduce_identity_fn, reduce_fn);
+    let best_map: Vec<usize> = best_result.map.iter().map(|x| x.index()).collect();
+    let mapping: HashMap<usize, usize> = best_map
+        .iter()
+        .enumerate()
+        .map(|(best_edge, edge)| (*edge, best_edge))
+        .collect();
+    let new_cmap: Vec<[usize; 2]> = best_result
+        .subgraph
+        .iter()
+        .map(|c| [mapping[&c[0].index()], mapping[&c[1].index()]])
+        .collect();
+    let rows: Vec<usize> = new_cmap.iter().map(|edge| edge[0]).collect();
+    let cols: Vec<usize> = new_cmap.iter().map(|edge| edge[1]).collect();
+    Ok((
+        rows.to_pyarray(py).into(),
+        cols.to_pyarray(py).into(),
+        best_map.to_pyarray(py).into(),
+    ))
+}
+
+/// Find densest subgraph in a :class:`~.PyGraph`
+///
+/// This method does not provide any guarantees on the approximation as it
+/// does a naive search using BFS traversal.
+///
+/// :param PyDigraph graph: The graph to find
+/// :param int num_nodes: The number of nodes in the subgraph to find
+/// :param func weight_callback: An optional callable that if specified will be
+///     passed the node indices of each edge in the graph and it is expected to
+///     return a float value. If specified the lowest avg weight for edges in
+///     a found subgraph will be a criteria for selection in addition to the
+///     connectivity of the subgraph.
+/// :returns: A tuple of 3 numpy arrays for efficient sparse matrix creation
+///     of the adjacency matrix of the subgraph mapping the values back to the
+///     node indices on the original graph.
+/// :rtype: (rows, cols, value)
+#[pyfunction]
+#[pyo3(text_signature = "(graph. num_nodes, /, weight_callback=None)")]
+pub fn graph_dense_subgraph(
+    py: Python,
+    graph: &graph::PyGraph,
+    num_nodes: usize,
+    weight_callback: Option<PyObject>,
+) -> PyResult<(PyObject, PyObject, PyObject)> {
+    densest_subgraph(py, &graph.graph, num_nodes, weight_callback)
+}
+
+/// Find densest subgraph in a :class:`~.PyDiGraph`
+///
+/// This method does not provide any guarantees on the approximation as it
+/// does a naive search using BFS traversal.
+///
+/// :param PyDigraph graph: The graph to find
+/// :param int num_nodes: The number of nodes in the subgraph to find
+/// :param func weight_callback: An optional callable that if specified will be
+///     passed the node indices of each edge in the graph and it is expected to
+///     return a float value. If specified the lowest avg weight for edges in
+///     a found subgraph will be a criteria for selection in addition to the
+///     connectivity of the subgraph.
+/// :returns: A tuple of 3 numpy arrays for efficient sparse matrix creation
+///     of the adjacency matrix of the subgraph mapping the values back to the
+///     node indices on the original graph.
+/// :rtype: (rows, cols, value)
+#[pyfunction]
+#[pyo3(text_signature = "(graph. num_nodes, /, weight_callback=None)")]
+pub fn digraph_dense_subgraph(
+    py: Python,
+    graph: &digraph::PyDiGraph,
+    num_nodes: usize,
+    weight_callback: Option<PyObject>,
+) -> PyResult<(PyObject, PyObject, PyObject)> {
+    densest_subgraph(py, &graph.graph, num_nodes, weight_callback)
+}

src/lib.rs

@@ -15,6 +15,7 @@ mod centrality;
 mod coloring;
 mod connectivity;
 mod dag_algo;
+mod dense_subgraph;
 mod digraph;
 mod dot_utils;
 mod generators;
@@ -38,6 +39,7 @@ use centrality::*;
 use coloring::*;
 use connectivity::*;
 use dag_algo::*;
+use dense_subgraph::*;
 use isomorphism::*;
 use layout::*;
 use matching::*;
@@ -405,6 +407,8 @@ fn retworkx(py: Python<'_>, m: &PyModule) -> PyResult<()> {
     m.add_wrapped(wrap_pyfunction!(articulation_points))?;
     m.add_wrapped(wrap_pyfunction!(biconnected_components))?;
     m.add_wrapped(wrap_pyfunction!(chain_decomposition))?;
+    m.add_wrapped(wrap_pyfunction!(digraph_dense_subgraph))?;
+    m.add_wrapped(wrap_pyfunction!(graph_dense_subgraph))?;
     m.add_class::<digraph::PyDiGraph>()?;
     m.add_class::<graph::PyGraph>()?;
     m.add_class::<toposort::TopologicalSorter>()?;