我正在尝试构建一个简单的模型,该模型可以将点分类为二维空间的 2 个分区:
- 我通过指定几个点和它们所属的分区来训练模型。
- 我使用该模型来预测组(分类)测试点可能落入。
不幸的是,我没有得到预期的答案。我是在我的代码中遗漏了什么还是我做错了什么?
public class SimpleClassifier {
public static class Point{
public int x;
public int y;
public Point(int x,int y){
this.x = x;
this.y = y;
}
@Override
public boolean equals(Object arg0) {
Point p = (Point) arg0;
return( (this.x == p.x) &&(this.y== p.y));
}
@Override
public String toString() {
// TODO Auto-generated method stub
return this.x + " , " + this.y ;
}
}
public static void main(String[] args) {
Map<Point,Integer> points = new HashMap<SimpleClassifier.Point, Integer>();
points.put(new Point(0,0), 0);
points.put(new Point(1,1), 0);
points.put(new Point(1,0), 0);
points.put(new Point(0,1), 0);
points.put(new Point(2,2), 0);
points.put(new Point(8,8), 1);
points.put(new Point(8,9), 1);
points.put(new Point(9,8), 1);
points.put(new Point(9,9), 1);
OnlineLogisticRegression learningAlgo = new OnlineLogisticRegression();
learningAlgo = new OnlineLogisticRegression(2, 2, new L1());
learningAlgo.learningRate(50);
//learningAlgo.alpha(1).stepOffset(1000);
System.out.println("training model \n" );
for(Point point : points.keySet()){
Vector v = getVector(point);
System.out.println(point + " belongs to " + points.get(point));
learningAlgo.train(points.get(point), v);
}
learningAlgo.close();
//now classify real data
Vector v = new RandomAccessSparseVector(2);
v.set(0, 0.5);
v.set(1, 0.5);
Vector r = learningAlgo.classifyFull(v);
System.out.println(r);
System.out.println("ans = " );
System.out.println("no of categories = " + learningAlgo.numCategories());
System.out.println("no of features = " + learningAlgo.numFeatures());
System.out.println("Probability of cluster 0 = " + r.get(0));
System.out.println("Probability of cluster 1 = " + r.get(1));
}
public static Vector getVector(Point point){
Vector v = new DenseVector(2);
v.set(0, point.x);
v.set(1, point.y);
return v;
}
}
输出:
ans =
no of categories = 2
no of features = 2
Probability of cluster 0 = 3.9580985042775296E-4
Probability of cluster 1 = 0.9996041901495722
99% 的输出显示 cluster 1
的概率更高。 为什么?
最佳答案
问题是您没有包含始终为 1 的偏差(截距)项。 您需要将偏差项 (1) 添加到您的点类中。
这是许多有机器学习经验的人犯的一个非常基本的错误。花一些时间学习理论可能是个好主意。 Andrew Ng's lectures是一个学习的好地方。
为了让您的代码给出预期的输出,需要更改以下内容。
- 添加了偏差项。
- 学习参数太高。改为 10
现在您将获得类 0 的 P(0)=0.9999。
这是一个给出正确结果的完整工作示例:
import java.util.HashMap;
import java.util.Map;
import org.apache.mahout.classifier.sgd.L1;
import org.apache.mahout.classifier.sgd.OnlineLogisticRegression;
import org.apache.mahout.math.DenseVector;
import org.apache.mahout.math.RandomAccessSparseVector;
import org.apache.mahout.math.Vector;
class Point{
public int x;
public int y;
public Point(int x,int y){
this.x = x;
this.y = y;
}
@Override
public boolean equals(Object arg0) {
Point p = (Point) arg0;
return( (this.x == p.x) &&(this.y== p.y));
}
@Override
public String toString() {
return this.x + " , " + this.y ;
}
}
public class SimpleClassifier {
public static void main(String[] args) {
Map<Point,Integer> points = new HashMap<Point, Integer>();
points.put(new Point(0,0), 0);
points.put(new Point(1,1), 0);
points.put(new Point(1,0), 0);
points.put(new Point(0,1), 0);
points.put(new Point(2,2), 0);
points.put(new Point(8,8), 1);
points.put(new Point(8,9), 1);
points.put(new Point(9,8), 1);
points.put(new Point(9,9), 1);
OnlineLogisticRegression learningAlgo = new OnlineLogisticRegression();
learningAlgo = new OnlineLogisticRegression(2, 3, new L1());
learningAlgo.lambda(0.1);
learningAlgo.learningRate(10);
System.out.println("training model \n" );
for(Point point : points.keySet()){
Vector v = getVector(point);
System.out.println(point + " belongs to " + points.get(point));
learningAlgo.train(points.get(point), v);
}
learningAlgo.close();
Vector v = new RandomAccessSparseVector(3);
v.set(0, 0.5);
v.set(1, 0.5);
v.set(2, 1);
Vector r = learningAlgo.classifyFull(v);
System.out.println(r);
System.out.println("ans = " );
System.out.println("no of categories = " + learningAlgo.numCategories());
System.out.println("no of features = " + learningAlgo.numFeatures());
System.out.println("Probability of cluster 0 = " + r.get(0));
System.out.println("Probability of cluster 1 = " + r.get(1));
}
public static Vector getVector(Point point){
Vector v = new DenseVector(3);
v.set(0, point.x);
v.set(1, point.y);
v.set(2, 1);
return v;
}
}
输出:
2 , 2 belongs to 0
1 , 0 belongs to 0
9 , 8 belongs to 1
8 , 8 belongs to 1
0 , 1 belongs to 0
0 , 0 belongs to 0
1 , 1 belongs to 0
9 , 9 belongs to 1
8 , 9 belongs to 1
{0:2.470723149516907E-6,1:0.9999975292768505}
ans =
no of categories = 2
no of features = 3
Probability of cluster 0 = 2.470723149516907E-6
Probability of cluster 1 = 0.9999975292768505
请注意,我在 SimpleClassifier 类之外定义了类 Point,但这只是为了使代码更具可读性,并不是必需的。
看看当你改变学习率时会发生什么。阅读有关交叉验证的说明以了解如何选择学习率。
Learning Rate => Probability of cluster 0
0.001 => 0.4991116089
0.01 => 0.492481585
0.1 => 0.469961472
1 => 0.5327745322
10 => 0.9745740393
100 => 0
1000 => 0
选择学习率:
- 运行随机梯度下降很常见,就像我们从固定学习率 α 开始所做的那样,慢慢地让学习率 α 降低到零 算法运行,也可以保证参数会收敛到 全局最小值,而不是仅仅围绕最小值振荡。
- 在这种情况下,当我们使用常量 α 时,您可以进行初始选择,运行梯度下降并观察成本函数,并相应地调整学习率。说明here
关于java - Mahout - 简单的分类问题,我们在Stack Overflow上找到一个类似的问题: https://stackoverflow.com/questions/11189456/