33-3-4/src/optimize-subcity.rs

use db::{MemoryLayoutDB, SqliteLayoutDB, LayoutDB, SavedLayout};
use city::{City, House, HouseLayout, get_price};
use itertools::Itertools;
use rand::{thread_rng, Rng, SeedableRng};
use rand::rngs::StdRng;
use rayon::prelude::*;

mod city;
mod db;
mod combine;
mod subcity;
mod optimization;
mod population;

fn main() {
    let mut sqlite_db = SqliteLayoutDB::from_file("layouts.sqlite").expect("Failed to load the DB");
    eprintln!("Loaded the DB, {} stored layouts", sqlite_db.layouts().len());

    let city = City::read_from_file("01.in", city::INPUT_CITY_WIDTH, city::INPUT_CITY_HEIGHT);
    eprintln!("Loaded the city file, {} houses", city.get_house_count());

    let best_layout: SavedLayout = sqlite_db.layouts().iter()
        .sorted_by(|x, y| get_price(&city, x.houses()).cmp(&get_price(&city, y.houses())))
        .map(|layout| (*layout).clone())
        .next().expect("No best layout found");
    eprintln!("Found best layout, ID {}, price {}", best_layout.id(), get_price(&city, best_layout.houses()));

    let x_range = 5533..=12000;
    let y_range = 4750..=12500;
    //let x_range = 5533..=8000;
    //let y_range = 4750..=8000;
    eprintln!("X {}-{}, Y {}-{}", x_range.start(), x_range.end(), y_range.start(), y_range.end());

    let static_houses: Vec<House> = best_layout.houses().iter()
        .filter(|house| !(x_range.contains(&house.x) && y_range.contains(&house.y)))
        .map(|&house| house)
        .collect();

    let removed_houses: Vec<House> = best_layout.houses().iter()
        .filter(|house| x_range.contains(&house.x) && y_range.contains(&house.y))
        .map(|&house| house)
        .collect();

    let removed_price: u32 = removed_houses.iter().map(|x| city.get_price(*x) as u32).sum();
    eprintln!("Price of all removed houses: {}", removed_price);

    let subcity = subcity::build_subcity(&city, &static_houses);

    eprintln!("Built subcity, width {}, height {}, {} houses, offset ({},{})",
              subcity.city().width(),
              subcity.city().height(),
              subcity.city().get_house_count(),
              subcity.x_offset(),
              subcity.y_offset()
    );

    //let mut subcity_db = MemoryLayoutDB::new();
    let filename = format!("X{}_{}Y{}_{}ID{}.sqlite", x_range.start(), x_range.end(), y_range.start(), y_range.end(), best_layout.id());
    let mut subcity_db = SqliteLayoutDB::from_file(&filename).unwrap();

    //for layout in subcity_db.layouts().iter()
    //    .filter(|x| get_price(subcity.city(), x.houses()) < removed_price)
    //    .sorted_by(|x, y| get_price(subcity.city(), x.houses()).cmp(&get_price(subcity.city(), y.houses()))) {
    //    let price = get_price(subcity.city(), layout.houses());
    //    let mut full_houses = subcity.to_full_houses(layout.houses());
    //    assert!(city::is_valid(&city, &full_houses).is_some());

    //    let mut house_layout = HouseLayout::new(&city);
    //    for house in &full_houses {
    //        house_layout.add_house(*house);
    //    }
    //    let seed: u64 = thread_rng().gen();
    //    let mut rng = StdRng::seed_from_u64(seed);
    //    optimization::iterate_improvements(&mut house_layout, &mut rng, true);
    //    eprintln!("Improvements finished");
    //    assert!(house_layout.is_valid());
    //    let improved_price = city::get_price(&city, house_layout.houses());
    //    if improved_price < city::get_price(&city, &full_houses) {
    //        eprintln!("Found improvement, new price {}, updating houses", improved_price);
    //        full_houses = house_layout.houses().clone();
    //    }

    //    assert!(city::is_valid(&city, &full_houses).is_some());
    //    println!("Layout {}, price {}, full price {}", layout.id(), price, city::get_price(&city, &full_houses));
    //    // Be careful with duplicates here
    //    //sqlite_db.add_layout(&full_houses, true);
    //    //println!("Inserted into the global DB");
    //}
    //return;


    // Architecture:
    // Build `FULL_RANDOM_LAYOUTS` random layouts
    // loop {
    //   Try combining `CUT_COMBINE_TOP_LAYOUTS` top layouts using vertical/horizontal cuts
    //   Generate `WEIGHTED_RANDOM_LAYOUTS` layouts weighted by scores from existing layouts
    // }

    const FULL_RANDOM_LAYOUTS: usize = 100;
    const DB_CHOICE_PROBABILITY: f64 = 0.90;
    const WEIGHTED_RANDOM_LAYOUTS: usize = 200;
    const CUT_COMBINE_TOP_LAYOUTS: usize = 500;
    const IGNORED_WEIGHT_RATIO: f64 = 0.5;

    if subcity_db.layouts().len() == 0 {
        let mut full_random_layouts = Vec::new();

        full_random_layouts.par_extend((0..FULL_RANDOM_LAYOUTS).into_par_iter().map(|i| {
            let seed: u64 = thread_rng().gen();
            let mut rng = StdRng::seed_from_u64(seed);
            let mut layout = HouseLayout::new(subcity.city());

            //eprintln!("Starting random population {}", i);
            population::populate_random(&mut layout, &mut rng);
            //eprintln!("Finished random init {}, price: {}, houses: {}", i, layout.price(), layout.houses().len());
            optimization::iterate_improvements(&mut layout, &mut rng, false);
            //eprintln!("Finished iterated improvements {}, price: {}, houses: {}", i, layout.price(), layout.houses().len());
            layout.houses().clone()
        }));

        for houses in &full_random_layouts {
            subcity_db.add_layout(houses, true);
        }

        eprintln!("Finished initial full random population");
    } else {
        eprintln!("Skipping initial full random population because DB is non-empty [{} layouts]", subcity_db.layouts().len());
    }

    let best_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).min().unwrap();
    let worst_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).max().unwrap();
    eprintln!("Best {}, worst {} [{} layouts]", best_price, worst_price, subcity_db.layouts().len());

    let mut cache = combine::CompatibilityCache::new();

    // TODO: Deduplication of the DB
    loop {
        // This is a bottleneck when it comes to multithreading, it only runs on a single thread
        combine::iterate_combines(&mut subcity_db, CUT_COMBINE_TOP_LAYOUTS, subcity.city(), &mut cache, false);

        let best_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).min().unwrap();
        let worst_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).max().unwrap();
        eprintln!("Finished cut combines");
        eprintln!("Best {}, worst {} [{} layouts]", best_price, worst_price, subcity_db.layouts().len());

        let mut weighted_random_layouts = Vec::new();
        // Only using the underlying memory-based DB is required here as the SqliteLayoutDB
        // is not thread-safe. We are only reading in the parallel iterator, so that's fine.
        let memory_db = subcity_db.memory_db();
        weighted_random_layouts.par_extend((0..WEIGHTED_RANDOM_LAYOUTS).into_par_iter().map(|i| {
            let seed: u64 = thread_rng().gen();
            let mut rng = StdRng::seed_from_u64(seed);
            let mut layout = HouseLayout::new(subcity.city());

            let price_range = worst_price - best_price;
            let max_price = best_price as f64 + (price_range as f64 * (1. - IGNORED_WEIGHT_RATIO));
            //eprintln!("Starting random weighted population {}, using DB, score range {}-{}, DB use probability {}...", i, best_price, worst_price, DB_CHOICE_PROBABILITY);
            population::populate_using_db(&mut layout, &mut rng, memory_db, best_price as f64, max_price as f64, DB_CHOICE_PROBABILITY);
            //eprintln!("Finished random init {}, price: {}, houses: {}", i, layout.price(), layout.houses().len());
            optimization::iterate_improvements(&mut layout, &mut rng, false);
            //eprintln!("Finished iterated improvements {}, price: {}, houses: {}", i, layout.price(), layout.houses().len());
            layout.houses().clone()
        }));

        for houses in &weighted_random_layouts {
            subcity_db.add_layout(houses, true);
        }

        let w_best: u32 = weighted_random_layouts.iter().map(|houses| houses.iter().map(|h| subcity.city().get_price(*h) as u32).sum()).min().unwrap();
        let w_worst: u32 = weighted_random_layouts.iter().map(|houses| houses.iter().map(|h| subcity.city().get_price(*h) as u32).sum()).max().unwrap();
        let best_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).min().unwrap();
        let worst_price: u32 = subcity_db.layouts().iter().map(|layout| layout.houses().iter().map(|h| subcity.city().get_price(*h) as u32).sum()).max().unwrap();
        eprintln!("Finished weighted random population, price range {}-{}", w_best, w_worst);
        eprintln!("Best {}, worst {} [{} layouts]", best_price, worst_price, subcity_db.layouts().len());
    }
}