2010 April

相較於前一篇所實作的KNN「分類」演算法，本篇所要實作的則是K-means「分群」演算法(Unsupervised learning)，兩者同樣都是使用Euclidean distance來決定和中心點的距離計算，只不過K-means還需要反覆不斷的修正K個中心點，直到K個點慢慢地趨於穩定才算完成，詳細的程式如下：

(本範例K值為3，所以一開始在畫面上點擊滑鼠左鍵會先依序分配3個分群的中心點，接著再不斷地點擊滑鼠左鍵就可以觀察到K點的變化，玩玩看吧>>kmeans)

Kmeans.as

package
{
	import flash.display.*;
	import flash.events.*;
	import flash.geom.Point;
	import flash.utils.*;
	
	public class Kmeans extends MovieClip
	{
		private var k:uint = 3;
		private var count:uint = 0;
		private var _r:uint = 0;
		private var _g:uint = 0;
		private var _b:uint = 0;
		private var k_center:Array = new Array();
		private var k_means:Array = new Array();
	
		public function Kmeans()
		{
			stage.addEventListener(MouseEvent.MOUSE_UP,put);
			init();
		}
		public function put(m:MouseEvent):void
		{
			if(count < k)
			{
				k_means[count] = new Array(); 
				var c:Circle = null;
				if(count == 0)
				{
					c = new Circle(0x0000FF);
				}else if(count == 1)
				{
					c = new Circle(0x00FF00);
				}else{
					c = new Circle(0xFF0000);
				}
				c.x = stage.mouseX;
				c.y = stage.mouseY;
				addChild(c);
				k_center[count] = c;
				count++;
				return;
			}
			for(var w = 0 ; w < 20 ; w++)
			{
				var cc2:Circle = this.getChildAt(w) as Circle;
				var k_arr:Array = new Array();
				var k_dict:Dictionary = new Dictionary(true);
				for(var v = 0 ; v < k ;v++)
				{
					var k_c = k_center[v];
					var dist2 = Point.distance(new Point(k_c.x,k_c.y),new Point(cc2.x,cc2.y));
					k_dict[dist2] = k_c;
					k_arr.push(dist2);
				}
				k_arr.sort(Array.NUMERIC);
				var min = k_arr[0];
				var color:uint = k_dict[min].getColor();
				cc2.reDraw(color);
				if(color == 0x0000FF)
				{
					k_means[0][_b] = new Point(cc2.x,cc2.y);
					_b++;
				}else if(color == 0x00FF00)
				{
					k_means[1][_g] = new Point(cc2.x,cc2.y);
					_g++;
				}else{
					k_means[2][_r] = new Point(cc2.x,cc2.y);
					_r++;
				}
			}
			trace(_r+" "+_g+" "+_b);
			var xx=0,yy=0;
			for(var bb = 0 ; bb < _b ; bb++)
			{
				xx += k_means[0][bb].x;
				yy += k_means[0][bb].y;
			}
			k_center[0].x = xx/_b;
			k_center[0].y = yy/_b;
			xx=0;yy=0;
			for(var gg = 0 ; gg < _g ; gg++)
			{
				xx += k_means[1][gg].x;
				yy += k_means[1][gg].y;
			}
			k_center[1].x = xx/_g;
			k_center[1].y = yy/_g;
			xx=0;yy=0;
			for(var rr = 0 ; rr < _r ; rr++)
			{
				xx += k_means[2][rr].x;
				yy += k_means[2][rr].y;
			}
			k_center[2].x = xx/_r;
			k_center[2].y = yy/_r;
			trace(k_center[0].x+" "+k_center[0].y);
		}
		public function init():void
		{
			for(var i = 0 ; i < 10 ; i++)
			{
				var c:Circle = new Circle(0x777777);				
				c.x = Math.random()*500;
				c.y = Math.random()*500;
				addChild(c);				
			}
			for(i = 0 ; i < 10 ; i++)
			{
				var c2:Circle = new Circle(0x777777);				
				c2.x = Math.random()*500;
				c2.y = Math.random()*500;
				addChild(c2);
			}
		}
	}
}

Circle.as

package
{
	import flash.display.*;
	
	public class Circle extends Sprite
	{
		private var color:uint = 0;
		
		public function Circle(c:uint)
		{
			this.color = c;
			this.graphics.lineStyle(5);
			this.graphics.beginFill(this.color);
			this.graphics.drawCircle(0,0,20);
			this.graphics.endFill();
		}
		public function getColor():uint
		{
			return this.color;
		}
		public function reDraw(color:uint):void
		{
			this.graphics.clear();
			this.graphics.lineStyle(5);
			this.graphics.beginFill(color);
			this.graphics.drawCircle(0,0,20);
			this.graphics.endFill();
			this.alpha = .5;
		}
	}
}

最近本來想打算藉由Flash CS5的發佈來開發iPhone應用程式，無奈Apple的大刀一揮~ 導致就算用Flash CS5開發出來的iPhone程式也通過不了上架的審核... 一下子打亂了我的一些計劃... 不過還是要練習練習AS3，而本文主要就是用AS3來實現KNN的基本概念，簡單來說KNN是一種可以藉由訓練資料來推測可能結果的演算法(Supervised learning)，詳細的程式細節如下：

(本範例初始化的K值為5，玩玩看吧>>knn.swf)

KNN.as

package
{
	import flash.display.*;
	import flash.events.*;
	import flash.geom.Point;
	import flash.utils.*;
	
	public class KNN extends MovieClip
	{
		private var k:uint = 5;
	
		public function KNN()
		{
			stage.addEventListener(MouseEvent.MOUSE_UP,put);
			init();
		}
		public function put(m:MouseEvent):void
		{
			var c:Circle = null;			
			var num:uint = this.numChildren;
			trace(num);
			var k_dict:Dictionary = new Dictionary(true);
			var k_arr:Array = new Array();
			for(var i = 0 ; i < num ; i++)
			{
				var cc:Circle = this.getChildAt(i) as Circle;
				var dist = Point.distance(new Point(stage.mouseX,stage.mouseY),new Point(cc.x,cc.y));
				k_dict[dist] = i;
				k_arr.push(dist);
			}
			k_arr.sort(Array.NUMERIC);
			var blue = 0;
			var red = 0;
			for(var j = 0 ; j < k ; j++)
			{
				var nid = k_dict[k_arr[j]];
				var c2:Circle = this.getChildAt(nid) as Circle;
				var cor:uint = c2.getColor();
				trace(nid+" "+k_arr[j]+" "+cor);
				if(cor == 0x0000FF)
					blue++;
				else
					red++;
			}
			if(blue > red)
				c = new Circle(0x0000FF);				
			else
				c = new Circle(0xFF0000);				
			
			c.x = stage.mouseX;
			c.y = stage.mouseY;
			addChild(c);
		}
		public function init():void
		{
			for(var i = 0 ; i < 10 ; i++)
			{
				var c:Circle = new Circle(0xff0000);				
				c.x = Math.random()*500;
				c.y = Math.random()*500;
				addChild(c);				
			}
			for(i = 0 ; i < 10 ; i++)
			{
				var c2:Circle = new Circle(0x0000FF);				
				c2.x = Math.random()*500;
				c2.y = Math.random()*500;
				addChild(c2);
			}
		}
	}
}

Circle.as

package
{
	import flash.display.*;
	
	public class Circle extends Sprite
	{
		private var color:uint = 0;
		
		public function Circle(c:uint)
		{
			this.color = c;
			this.graphics.lineStyle(5);
			this.graphics.beginFill(this.color);
			this.graphics.drawCircle(0,0,20);
			this.graphics.endFill();
		}
		public function getColor():uint
		{
			return this.color;
		}
	}
}

相關資源

．KNN演算法

．KNN演算法 (更正篇)

有一陣子沒po些東西了... 先來放一張之前去「白沙灣」所拍的螃蟹特寫充充數 XD