Files
algos/src/k_clustering_big.rs

64 lines
2.3 KiB
Rust

use std::collections::HashMap;
pub type ImpliciteGraph = Vec<u32>;
fn neighbors(a: u32, n: usize) -> Vec<u32> {
let mut r = vec![a];
for _ in 0..n {
let mut r_new = r.clone();
for a in r {
for i in 0..24 {
let m = 0x1 << i;
let neighbor = a ^ m;
r_new.push(neighbor);
}
}
r_new.sort();
r_new.dedup();
r = r_new;
}
r
}
pub fn k_clustering_big(g: &ImpliciteGraph) -> usize {
let mut node_id_to_cluster_id: Vec<usize> = (0..g.len()).collect();
let mut clusters: Vec<Vec<usize>> = (0..g.len()).map(|x| vec![x]).collect();
let mut node_map: HashMap<u32, Vec<usize>> = HashMap::new();
let mut cluster_count = g.len();
for i in 0..g.len() {
if let Some(x) = node_map.get_mut(&g[i]) {
x.push(i);
} else {
node_map.insert(g[i], vec![i]);
}
}
for node_a_id in 0..g.len() {
// Iterate over all nodes in the graph. Then, for each node compute all
// neighbors that are two or less bits away.
for node_b_value in neighbors(g[node_a_id], 2) {
// See if there exist nodes that match the neighbor. If such nodes
// exist iterate over them and merge the clusters if they are not
// already the same. The key insight is that we have to cluster all
// nodes that are two or less (that includes zero) bits apart.
if let Some(node_b_ids) = node_map.get(&node_b_value) {
for node_b_id in node_b_ids {
let cluster_id_a = node_id_to_cluster_id[node_a_id];
let cluster_id_b = node_id_to_cluster_id[*node_b_id];
if cluster_id_a != cluster_id_b {
// Merge b into a. The code is the same as for k_clustering.
let mut cluster_b = std::mem::take(&mut clusters[cluster_id_b]);
for node_id in &cluster_b {
node_id_to_cluster_id[*node_id] = cluster_id_a;
}
clusters[cluster_id_a].append(&mut cluster_b);
cluster_count -= 1;
}
}
}
}
}
cluster_count
}