- iOS/Objective-C 元类和类别
- objective-c - -1001 错误,当 NSURLSession 通过 httpproxy 和/etc/hosts
- java - 使用网络类获取 url 地址
- ios - 推送通知中不播放声音
我想要一个网格这意味着,给定一个n x m的网格,其中有黄色和红色的正方形,我希望以这样的方式对网格进行分区,即黄色将是尽可能多的分区中的大多数颜色,如下图所示:
所有的分区必须是连续的,相同数量的正方形,所有的正方形都是彩色的(尽管如果一个算法可以推广到一些没有彩色的正方形的网格上,那就太棒了)。
我不知道如何去“算法化”这个问题,过去的暴力迫使每一个可能的分区,这本身就足够困难,是难以置信的低效。
最好的方法是什么?
最佳答案
tl;dr:使用模拟退火,在地区之间交换选民。
底部的演示程序允许您执行交换步骤(随机演化)并优化gerrymandered区域(anneal)
一般优化
我们可以把这个框架化为最优化问题,我们试图最大化地区的数量。
红队获胜,蓝队获胜的地区数量也减少了。
让我们把这个形式化:
function heuristic(state) {
return state.districts_that_red_wins - state.districts_that_blue_wins;
}
state
是选民分配到地区。
function heuristic(state) {
let {red_wins, blue_wins, tied, wasted_red_votes, wasted_blue_votes} = get_winners(state, voters);
return red_wins - blue_wins - 0.1 * wasted_red_votes + 0.1 * wasted_blue_votes;
}
state
。
max(0.8 - iterations/total_iterations, 0)
。在开始的时候,20%的时间,我们会采取一个新的状态,如果而且只有当它更好;其他80%的时间,我们会采取新的状态,无论。
anneal
函数,或者直接阅读wikipedia文章。
is_swappable
函数。这就是演示中的“随机进化”按钮所做的。
function evolve_state() {
randomly pick a source district
randomly pick a non-articulation point, source_point, from source_district
for each neighbour of the articulation point
if the neighbour is in a different district target_district
temporarily remove source_point from source_district and add it to target_district
if any articulation point (other than source point), target_point, in target_district is adjacent to source_district
swap target_point and source_point
return;
restore source_point
}
source_district
、
source_point
、
neighbour
、
target_district
和
target_point
的方式实现了这一点,因为我不确定这会有多稀疏。
evolve_state
。
const RED = 'R';
const BLUE = 'B';
const VOTERS_PER_DISTRICT = 10;
const neighbours = [{x: 1, y: 0}, {x: 0, y: 1}, {x: -1, y: 0}, {x: 0, y: -1}];
/* UTILITY FUNCTIONS */
/**
Create a generator that starts at a random point p, 0 <= p < max
The generator will iterate over values p, p+1, ... max, 0, ... p-1
*/
function* cyclic_generator(max) {
let start = _.random(max);
for (let i=0; i<max; i++) {
yield (i + start) % max;
}
}
/**
Return grid[x][y] if x and y are within grid. Otherwise return undefined
*/
function grid_get(grid, x, y) {
if(_.isUndefined(grid[x])) {
return undefined;
}
else {
return grid[x][y];
}
}
/** Generates a 2d array red and blue voters */
function generate_voters() {
return _.times(5, x => _.times(10, () => {return {vote: x > 2 ? RED : BLUE, district_vote: 0xffffff}}))
}
/** Generate an initial state */
function generate_initial_state() {
return _.range(5).map(x => _.range(10).map(y => {return {x, y}}));
}
/**
Randomly swap two squares in the grid between two districts.
The new square to be added must be connected to the district, and the
old square must not break another district in two
*/
function evolve_state(state) {
state = _.cloneDeep(state);
// Create a grid with the district number
let point_to_district = _.range(5).map(x => _.range(10).map(y => -1));
state.forEach((district, i) => district.forEach(({x, y}) => point_to_district[x][y] = i));
// swap a point from source_district to target_district.
// then swap a point from target_district to source_district.
for(let source_district_idx of cyclic_generator(state.length)) {
let source_articulation_points = state[source_district_idx].filter(point => is_swappable(point_to_district, point, source_district_idx));
for(let source_point_idx of cyclic_generator(source_articulation_points.length)) {
let source_point = source_articulation_points[source_point_idx];
for(let neighbour_idx of cyclic_generator(4)) {
let neighbour = neighbours[neighbour_idx];
let target_district_idx = grid_get(point_to_district, source_point.x + neighbour.x, source_point.y + neighbour.y);
if (_.isUndefined(target_district_idx) || target_district_idx == source_district_idx) {
continue;
}
// swap the source point
point_to_district[source_point.x][source_point.y] = target_district_idx;
_.remove(state[source_district_idx], ({x, y}) => x == source_point.x && y == source_point.y);
// we don't add the point the the target array yet because we don't want to swap that point back
// try to find a point in target_district that we can move to source_district
let target_articulation_points = state[target_district_idx].filter(point => is_swappable(point_to_district, point, target_district_idx));
for(let target_point_idx of cyclic_generator(target_articulation_points.length)) {
let target_point = target_articulation_points[target_point_idx];
for(let n of neighbours) {
if(grid_get(point_to_district, target_point.x + n.x, target_point.y + n.y) === source_district_idx) {
// found a point that we can swap!
// console.log('swapping points!', source_point, target_point);
_.remove(state[target_district_idx], ({x, y}) => x == target_point.x && y == target_point.y);
state[target_district_idx].push(source_point);
state[source_district_idx].push(target_point);
return state;
}
}
}
// unswap source point since we were unsuccessful
point_to_district[source_point.x][source_point.y] = source_district_idx;
state[source_district_idx].push(source_point);
}
}
}
throw 'Could not find any states to swap' // this should never happen, since there will always be the option of reversing the previous step
}
/*
Return whether a point can be removed from a district without creating disjoint districts.
In graph theory, points that cannot be removed are articulation points.
For a general algorithm, see: https://stackoverflow.com/questions/15873153/explanation-of-algorithm-for-finding-articulation-points-or-cut-vertices-of-a-gr
My version takes advantage of the fact that we're on a grid and that all the districts must be continuous,
so we can consider only the immediate neighbours of a point.
*/
function is_swappable(grid, p, district) {
// if the the point is not even in this district, it makes no sense for this to consider this point at all
if(grid[p.x][p.y] != district) {
return false;
}
// if two opposite edges are part of this district, this is an articulation point
// .x. x is an articulation point
// Exception:
// .x. x is not an articulation point
// ...
if (grid_get(grid, p.x+1, p.y) === district && grid_get(grid, p.x-1, p.y) === district && grid_get(grid, p.x, p.y+1) !== district && grid_get(grid, p.x, p.y-1) !== district) {
return false;
}
if (grid_get(grid, p.x, p.y+1) === district && grid_get(grid, p.x, p.y-1) === district && grid_get(grid, p.x+1, p.y) !== district && grid_get(grid, p.x-1, p.y) !== district) {
return false;
}
// check if any corners are missing:
// .x x is not an articulation point .x x is an articulation point
// .. .
for(let i = 0; i < 4; i++) {
let nx = neighbours[i].x;
let ny = neighbours[i].y;
let nx2 = neighbours[(i+1)%4].x;
let ny2 = neighbours[(i+1)%4].y;
if (grid_get(grid, p.x+nx, p.y+ny) === district && grid_get(grid, p.x+nx2, p.y+ny2) === district && grid_get(grid, p.x+nx+nx2, p.y+ny+ny2) !== district) {
return false;
}
}
return true;
}
/** Count how many districts each party wins */
function get_winners(state, voters) {
let red_wins = 0;
let blue_wins = 0;
let tied = 0;
let wasted_red_votes= 0; // see https://en.wikipedia.org/wiki/Wasted_vote
let wasted_blue_votes = 0;
state.forEach(district => {
let counts = _.countBy(district.map(({x, y}) => voters[x][y].vote))
if ((counts[BLUE] || 0) > (counts[RED] || 0)) {
blue_wins++;
wasted_blue_votes += (counts[BLUE] || 0) - VOTERS_PER_DISTRICT / 2 - 1;
wasted_red_votes += (counts[RED] || 0);
}
else if ((counts[RED] || 0) > (counts[BLUE] || 0)) {
red_wins++;
wasted_red_votes += (counts[RED] || 0) - VOTERS_PER_DISTRICT / 2 - 1;
wasted_blue_votes += (counts[BLUE] || 0);
}
else {
tied++;
}
});
return {red_wins, blue_wins, tied, wasted_red_votes, wasted_blue_votes};
}
/* GUI */
/* Display a grid showing which districts each party won */
function render_districts(state, voters) {
let red_districts = 0;
let blue_districts = 0;
let grey_districts = 0;
// Color each district
state.forEach(district => {
let counts = _.countBy(district.map(({x, y}) => voters[x][y].vote))
let district_color;
if ((counts[BLUE] || 0) > (counts[RED] || 0)) {
district_color = 'blue' + blue_districts++;
}
else if ((counts[RED] || 0) > (counts[BLUE] || 0)) {
district_color = 'red' + red_districts++;
}
else {
district_color = 'grey' + grey_districts++;
}
district.map(({x, y}) => voters[x][y].district_color = district_color);
});
return m('table', [
m('tbody', voters.map(row =>
m('tr', row.map(cell => m('td', {'class': cell.district_color}, cell.vote)))
))
]);
}
/** Score a state with four criteria:
- maximize number of red districts
- minimize number of blue districts
- minimize number of red voters in districts that red wins
- maximize number of blue voters in districts that blue wins
The first two criteria are arbitrarily worth 10x more than the latter two
The latter two are to nudge the final result toward the correct solution
*/
function heuristic(state) {
let {red_wins, blue_wins, tied, wasted_red_votes, wasted_blue_votes} = get_winners(state, voters);
return red_wins - blue_wins - 0.1 * wasted_red_votes + 0.1 * wasted_blue_votes;
}
/**
Optimization routine to find the maximum of prob_fcn.
prob_fcn: function to maximize. should take state as its argument
transition: how to generate another state from the previous state
initialize_state: a function that returns an initial state
iters: number of iterations to run
Stolen from my repo here: https://github.com/c2huc2hu/automated-cryptanalysis/blob/master/part3.js
*/
function anneal(prob_fcn, transition, initialize_state, seeds=1, iters=1000) {
let best_result = initialize_state();
for(let i=0; i<seeds; i++) {
let curr_state = initialize_state();
let curr_cost = prob_fcn(curr_state);
// perform annealing. do a few extra steps with temp=0 to refine the final solution
for(let j=0; j<iters; j++) {
let candidate_state = transition(curr_state);
let candidate_cost = prob_fcn(candidate_state);
temp = 0.8 - j / iters;
if(candidate_cost >= curr_cost || Math.random() < temp) {
curr_state = candidate_state;
curr_cost = candidate_cost;
}
}
if(prob_fcn(curr_state) > prob_fcn(best_result)) {
best_result = curr_state;
}
}
return best_result;
}
let voters = generate_voters();
let state = generate_initial_state();
// main rendering code: this code renders the UI
m.mount(document.getElementById('actions'), {view: function() {
return m('div', [
m('button', {onclick: () => state = generate_initial_state()}, 'Reset'),
m('button', {onclick: () => state = evolve_state(state)}, 'Randomly evolve'), // randomly evolves
m('br'),
m('label', {'for': 'radio-blue'}, 'Gerrymander for blue'),
m('input', {type: 'radio', name: 'heuristic', value: 'blue', id: 'radio-blue'}),
m('label', {'for': 'radio-red'}, 'Gerrymander for red'),
m('input', {type: 'radio', name: 'heuristic', value: 'red', id: 'radio-red'}),
m('br'),
m('label', {'for': 'anneal-steps'}, 'Anneal steps: '),
m('input', {id: 'anneal-steps', type: 'number', value: '500'}),
m('button', {onclick: function() {
let minimize = document.getElementById('radio-red').checked;
let _heuristic = minimize ? heuristic : state => -heuristic(state)
let new_state = anneal(_heuristic, evolve_state, generate_initial_state, 1, parseInt(document.getElementById('anneal-steps').value));
if(_heuristic(new_state) > _heuristic(state)) {
state = new_state;
}
else {
console.log('found no better solutions')
}
}}, 'Anneal!'),
]);
}});
// This renders the grid
m.mount(document.getElementById('grid'), {
view: function() {
return render_districts(state, voters)
}
});
// state = anneal(heuristic, evolve_state, generate_initial_state, 5, 1000);
document.getElementById('radio-red').checked = true;
m.redraw();
/* Layout */
table {
border: solid 1px black;
}
td {
padding: 5px;
border: solid 1px black;
}
button {
margin: 10px;
}
p {
max-width: 500px;
}
/* Colour classes. In hindsight, this wasn't a good idea */
.red0 {
background-color: red;
}
.red1 {
background-color: darkred;
}
.red2 {
background-color: pink;
}
.red3 {
background-color: deeppink;
}
.red4 {
background-color: lightsalmon;
}
.blue0 {
background-color: aqua;
}
.blue1 {
background-color: cadetblue;
}
.blue2 {
background-color: steelblue;
}
.blue3 {
background-color: royalblue;
}
.blue4 {
background-color: midnightblue;
}
.grey0 {
background-color: lightgrey;
}
.grey1 {
background-color: silver;
}
.grey2 {
background-color: darkgray;
}
.grey3 {
background-color: gray;
}
.grey4 {
background-color: dimgray;
}
<!DOCTYPE html>
<html>
<head>
<script data-require="lodash.js@4.17.4" data-semver="4.17.4" src="https://cdn.jsdelivr.net/npm/lodash@4.17.4/lodash.min.js"></script>
<script data-require="mithril@1.0.1" data-semver="1.0.1" src="https://cdnjs.cloudflare.com/ajax/libs/mithril/1.0.1/mithril.js"></script>
<link rel="stylesheet" href="style.css" />
</head>
<body>
<h1>Gerrymandering simulation</h1>
<p>
There are two parties, red and blue (chosen because they contrast well).
Each person will always vote a certain way, and is marked with R or B in the table.
People are divided into districts, shown here as groups of people marked in a single colour.
</p>
<p>
Use the buttons below to divide up districts.
The reset button will restore the initial state.
The randomly-evolve button will swap two people between districts
The anneal button will optimize for your chosen party.
You should limit the number of steps to ~1000 or your browser will appear to hang.
In general, it is sufficient to run a few seeds for 500 iterations.
</p>
<div id="grid"></div>
<div id="actions"></div>
<script src="script.js"></script>
</body>
</html>
关于algorithm - 分区彩色网格,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/51640083/
我在一本书(Interview Question)中读到这个问题,想在这里详细讨论这个问题。请点亮它。 问题如下:- 隐私和匿名化 马萨诸塞州集团保险委员会早在 1990 年代中期就有一个绝妙的主意
我最近接受了一次面试,面试官给了我一些伪代码并提出了相关问题。不幸的是,由于准备不足,我无法回答他的问题。由于时间关系,我无法向他请教该问题的解决方案。如果有人可以指导我并帮助我理解问题,以便我可以改
这是我的代码 public int getDist(Node root, int value) { if (root == null && value !=0) return
就效率而言,Strassen 算法应该停止递归并应用乘法的最佳交叉点是多少? 我知道这与具体的实现和硬件密切相关,但对于一般情况应该有某种指南或某人的一些实验结果。 在网上搜索了一下,问了一些他们认为
我想学习一些关于分布式算法的知识,所以我正在寻找任何书籍推荐。我对理论书籍更感兴趣,因为实现只是个人喜好问题(我可能会使用 erlang(或 c#))。但另一方面,我不想对算法进行原始的数学分析。只是
我想知道你们中有多少人实现了计算机科学的“ classical algorithms ”,例如 Dijkstra's algorithm或现实世界中的数据结构(例如二叉搜索树),而不是学术项目? 当有
我正在解决旧编程竞赛中的一些示例问题。在这个问题中,我们得到了我们有多少调酒师以及他们知道哪些食谱的信息。制作每杯鸡尾酒需要 1 分钟,我们需要使用所有调酒师计算是否可以在 5 分钟内完成订单。 解决
关闭。这个问题是opinion-based .它目前不接受答案。 想要改进这个问题? 更新问题,以便 editing this post 可以用事实和引用来回答它. 关闭 8 年前。 Improve
我开始学习 Nodejs,但我被困在中间的某个地方。我从 npm 安装了一个新库,它是 express -jwt ,它在运行后显示某种错误。附上代码和错误日志,请帮助我! const jwt = re
我有一个证书,其中签名算法显示“sha256rsa”,但指纹算法显示“sha1”。我的证书 SHA1/SHA2 的标识是什么? 谢谢! 最佳答案 TL;TR:签名和指纹是完全不同的东西。对于证书的强度
我目前在我的大学学习数据结构类(class),并且在之前的类(class)中做过一些算法分析,但这是我在之前的类(class)中遇到的最困难的部分。我们现在将在我的数据结构类(class)中学习算法分
有一个由 N 个 1x1 方格组成的区域,并且该区域的所有部分都是相连的(没有任何方格无法到达的方格)。 下面是一些面积的例子。 我想在这个区域中选择一些方块,并且两个相邻的方块不能一起选择(对角接触
我有一些多边形形状的点列表,我想将其包含在我页面上的 Google map 中。 我已经从原始数据中删除了尽可能多的不必要的多边形,现在我剩下大约 12 个,但它们非常详细以至于导致了问题。现在我的文
我目前正在实现 Marching Squares用于计算等高线曲线,我对此处提到的位移位的使用有疑问 Compose the 4 bits at the corners of the cell to
我正在尝试针对给定算法的约束满足问题实现此递归回溯函数: function BACKTRACKING-SEARCH(csp) returns solution/failure return R
是否有包含反函数的库? 作为项目的一部分,我目前正在研究测向算法。我正在使用巴特利特相关性。在 Bartlett 相关性中,我需要将已经是 3 次矩阵乘法(包括 Hermitian 转置)的分子除以作
关闭。这个问题不符合Stack Overflow guidelines .它目前不接受答案。 这个问题似乎与 help center 中定义的范围内的编程无关。 . 关闭 8 年前。 Improve
问题的链接是UVA - 1394 : And There Was One . 朴素的算法是扫描整个数组并在每次迭代中标记第 k 个元素并在最后停止:这需要 O(n^2) 时间。 我搜索了一种替代算法并
COM 中创建 GUID 的函数 (CoCreateGUID) 使用“分散唯一性算法”,但我的问题是,它是什么? 谁能解释一下? 最佳答案 一种生成 ID 的方法,该 ID 具有一定的唯一性保证,而不
在做一个项目时我遇到了这个问题,我将在这个问题的实际领域之外重新措辞(我想我可以谈论烟花的口径和形状,但这会使理解更加复杂).我正在寻找一种(可能是近似的)算法来解决它。 我有 n 个不同大小的容器,
我是一名优秀的程序员,十分优秀!