move model into own struct

This commit is contained in:
Niko Abeler 2022-11-18 19:32:13 +01:00
parent d91e829bff
commit 22a46b5c68
3 changed files with 103 additions and 63 deletions

12
src/graph.rs Normal file
View File

@ -0,0 +1,12 @@
use std::sync::{Arc, RwLock};
pub struct Node {
pub x: f32,
pub y: f32,
}
pub struct Edge {
pub weight: f32,
}
pub type EdgeMatrix = Arc<RwLock<Vec<Vec<Edge>>>>;
pub type NodeVector = Arc<Vec<RwLock<Node>>>;

View File

@ -1,28 +1,21 @@
mod utils;
mod spring_model;
mod graph;
use rand::Rng; use rand::Rng;
use std::fs::File; use std::fs::File;
use std::io::prelude::*; use std::io::prelude::*;
use std::sync::{Arc, RwLock}; use std::sync::{Arc, RwLock};
use std::thread; use std::thread;
use graph::{EdgeMatrix, NodeVector, Node, Edge};
mod utils;
struct Node { fn nodes_list(size: usize) -> NodeVector {
x: f32,
y: f32,
}
struct Edge {
weight: f32,
}
type EdgeMatrix = Arc<RwLock<Vec<Vec<Edge>>>>;
fn nodes_list(size: usize) -> Arc<Vec<RwLock<Node>>> {
let mut nodes = Vec::new(); let mut nodes = Vec::new();
for _ in 0..size { for _ in 0..size {
let node = RwLock::new(Node { let node = RwLock::new(Node {
x: rand::thread_rng().gen_range(0.0..100.0), x: rand::thread_rng().gen_range(-0.5..0.5),
y: rand::thread_rng().gen_range(0.0..100.0), y: rand::thread_rng().gen_range(-0.5..0.5),
}); });
nodes.push(node); nodes.push(node);
} }
@ -69,73 +62,32 @@ fn read_graph(file_name: &str) -> (usize, EdgeMatrix) {
} }
fn main() -> std::io::Result<()> { fn main() -> std::io::Result<()> {
const C_REP: f32 = 0.1;
const C_SPRING: f32 = 0.1;
const ITER: usize = 200; const ITER: usize = 200;
const THREADS: usize = 8; const THREADS: usize = 8;
// let edges = connection_matrix(size); // let edges = connection_matrix(size);
let (size, edges): (usize, EdgeMatrix) = read_graph("../graph.bin"); let (size, edges): (usize, EdgeMatrix) = read_graph("../debug_graph.bin");
println!("Size: {}", size); println!("Size: {}", size);
let nodes = nodes_list(size); let nodes = nodes_list(size);
let nodes_next = nodes_list(size); let nodes_next = nodes_list(size);
let opt_dist = 10.0 / (size as f32).sqrt(); let model = Arc::new(RwLock::new(spring_model::InitialModel::new(edges, size)));
let chunks = utils::gen_chunks(size, THREADS); let chunks = utils::gen_chunks(size, THREADS);
for epoch in 0..ITER { for epoch in 0..ITER {
model.write().unwrap().prepare(&nodes);
let mut handles = vec![]; let mut handles = vec![];
for i in 0..THREADS { for i in 0..THREADS {
let nodes = nodes.clone(); let nodes = nodes.clone();
let nodes_next = nodes_next.clone(); let nodes_next = nodes_next.clone();
let edges = edges.clone();
let chunk = chunks[i].clone(); let chunk = chunks[i].clone();
let model = model.clone();
let handle = thread::spawn(move || { let handle = thread::spawn(move || {
for n in chunk { for n in chunk {
let node = nodes[n].read().unwrap(); let update = model.read().unwrap().step(&nodes,n);
let edges = edges.read().unwrap();
let mut node_x = node.x;
let mut node_y = node.y;
for o in 0..size {
if o == n {
continue;
}
let o_x: f32;
let o_y: f32;
{
let other = nodes[o].read().unwrap();
o_x = other.x;
o_y = other.y;
}
let d_x = o_x - node_x;
let d_y = o_y - node_y;
let dist = (d_x * d_x + d_y * d_y).sqrt().max(0.0001);
let unit_x = d_x / dist;
let unit_y = d_y / dist;
let edge = edges[n][o].weight;
if edge == 0.0 {
let f_rep = -C_REP / (dist).powi(2);
let f_rep_x = f_rep * unit_x;
let f_rep_y = f_rep * unit_y;
node_x += f_rep_x;
node_y += f_rep_y;
} else {
let f_spring = C_SPRING * (dist / opt_dist).log(2.0);
let f_spring_x = f_spring * unit_x;
let f_spring_y = f_spring * unit_y;
node_x += f_spring_x;
node_y += f_spring_y;
}
}
let mut result = nodes_next[n].write().unwrap(); let mut result = nodes_next[n].write().unwrap();
result.x = node_x; result.x = update.x;
result.y = node_y; result.y = update.y;
} }
}); });
handles.push(handle); handles.push(handle);

View File

@ -0,0 +1,76 @@
use crate::graph::{EdgeMatrix, NodeVector, Node};
const C_REP: f32 = 0.1;
const C_SPRING: f32 = 0.1;
pub struct InitialModel {
edges: EdgeMatrix,
size: usize,
opt_dist: f32,
t: f32,
}
impl InitialModel {
pub fn new(edges: EdgeMatrix, size: usize) -> InitialModel {
let opt_dist = 1.0 / (size as f32).sqrt();
InitialModel{ edges, size, opt_dist, t: 0.1 }
}
pub fn prepare(& mut self, nodes: &NodeVector) {
self.t = 0.1 * f32::max(
nodes.iter().map(|n| n.read().unwrap().x.abs()).reduce(|a, b| a.max(b)).unwrap(),
nodes.iter().map(|n| n.read().unwrap().y.abs()).reduce(|a, b| a.max(b)).unwrap()
);
}
pub fn step(&self, nodes: &NodeVector, i_node: usize) -> Node {
let node = nodes[i_node].read().unwrap();
let edges = self.edges.read().unwrap();
let mut node_x = node.x;
let mut node_y = node.y;
for o in 0..self.size {
if o == i_node {
continue;
}
let o_x: f32;
let o_y: f32;
{
let other = nodes[o].read().unwrap();
o_x = other.x;
o_y = other.y;
}
let d_x = o_x - node_x;
let d_y = o_y - node_y;
let dist = (d_x * d_x + d_y * d_y).sqrt().max(0.01);
let unit_x = d_x / dist;
let unit_y = d_y / dist;
let edge = edges[i_node][o].weight;
if edge == 0.0 {
let f_rep = (C_REP / (dist).powi(2)).min(self.t);
let f_rep_x = f_rep * unit_x;
let f_rep_y = f_rep * unit_y;
node_x -= f_rep_x;
node_y -= f_rep_y;
} else {
let f_spring = (C_SPRING * (dist / self.opt_dist).log(2.0)).min(self.t);
let f_spring_x = f_spring * unit_x;
let f_spring_y = f_spring * unit_y;
node_x += f_spring_x;
node_y += f_spring_y;
}
}
Node {
x: node_x,
y: node_y,
}
}
}