机器学习的1NN最近邻算法,在weka里叫IB1,是因为Instance Base 1 ,也就是只基于一个最近邻的实例的惰性学习算法。
下面总结一下,weka中对IB1源码的学习总结。
首先需要把 weka-src.jar 引入编译路径,否则无法跟踪源码。
1)读取data数据,完成 IB1 分类器的调用,结果预测评估。为了后面的跟踪。
try { File file = new File("F:\tools/lib/data/contact-lenses.arff"); ArffLoader loader = new ArffLoader(); loader.setFile(file); ins = loader.getDataSet(); // 在使用样本之前一定要首先设置instances的classIndex,否则在使用instances对象是会抛出异常 ins.setClassIndex(ins.numAttributes() - 1); cfs = new IB1(); cfs.buildClassifier(ins); Instance testInst; Evaluation testingEvaluation = new Evaluation(ins); int length = ins.numInstances(); for (int i = 0; i < length; i++) { testInst = ins.instance(i); // 通过这个方法来用每个测试样本测试分类器的效果 double predictValue = cfs.classifyInstance(testInst); System.out.println(testInst.classValue()+"--"+predictValue); } // System.out.println("分类器的正确率:" + (1 - testingEvaluation.errorRate())); } catch (Exception e) { e.printStackTrace(); }
2)ctrl 点击buildClassifier,进一步跟踪buildClassifier方法的源码,在IB1的类中重写了这个抽象方法,源码为:
public void buildClassifier(Instances instances) throws Exception { // can classifier handle the data? getCapabilities().testWithFail(instances); // remove instances with missing class instances = new Instances(instances); instances.deleteWithMissingClass(); m_Train = new Instances(instances, 0, instances.numInstances()); m_MinArray = new double [m_Train.numAttributes()]; m_MaxArray = new double [m_Train.numAttributes()]; for (int i = 0; i < m_Train.numAttributes(); i++) { m_MinArray[i] = m_MaxArray[i] = Double.NaN; } Enumeration enu = m_Train.enumerateInstances(); while (enu.hasMoreElements()) { updateMinMax((Instance) enu.nextElement()); } }
(1)if是判断,IB1分类器不能处理属性是字符串和类别是数值型的样本;
(2)if是判断,删除没有类标签的样本;
(3)m_MinArray 和 m_MaxArray 分别保存最小和最大值,并且初始化double数组【样本个数】;
(4)遍历所有的训练样本实例,求最小和最大值;继续跟踪updateMinMax方法;
3)IB1类的updateMinMax方法的源码如下:
private void updateMinMax(Instance instance) { for (int j = 0;j < m_Train.numAttributes(); j++) { if ((m_Train.attribute(j).isNumeric()) && (!instance.isMissing(j))) { if (Double.isNaN(m_MinArray[j])) { m_MinArray[j] = instance.value(j); m_MaxArray[j] = instance.value(j); } else { if (instance.value(j) < m_MinArray[j]) { m_MinArray[j] = instance.value(j); } else { if (instance.value(j) > m_MaxArray[j]) { m_MaxArray[j] = instance.value(j); } } } } } }
(1)过滤掉属性不是数值型和缺失标签的实例;
(2)若是isNaN,is not a number,是数值型的话,循环遍历样本的每一个属性,求出最大最小值;
到此为止,训练了IB1模型(有人可能会问lazy的算法难道不是不需要训练模型吗?我认为build分类器是为了初始化 m_Train和求所有实例的每个属性的最大最小值,为了下一步求distance做准备)
下面介绍下预测源码:
4)跟踪classifyInstance方法,源码如下:
public double classifyInstance(Instance instance) throws Exception { if (m_Train.numInstances() == 0) { throw new Exception("No training instances!"); } double distance, minDistance = Double.MAX_VALUE, classValue = 0; updateMinMax(instance); Enumeration enu = m_Train.enumerateInstances(); while (enu.hasMoreElements()) { Instance trainInstance = (Instance) enu.nextElement(); if (!trainInstance.classIsMissing()) { distance = distance(instance, trainInstance); if (distance < minDistance) { minDistance = distance; classValue = trainInstance.classValue(); } } } return classValue; }
(1)调用方法updateMinMax更新了加入测试实例后的最大最小值;
(2)计算测试实例到每一个训练实例的距离,distance方法,并且保存距离最小的实例minDistance;
5)跟踪classifyInstance方法,源码如下:
private double distance(Instance first, Instance second) { double diff, distance = 0; for(int i = 0; i < m_Train.numAttributes(); i++) { if (i == m_Train.classIndex()) { continue; } if (m_Train.attribute(i).isNominal()) { // If attribute is nominal if (first.isMissing(i) || second.isMissing(i) || ((int)first.value(i) != (int)second.value(i))) { distance += 1; } } else { // If attribute is numeric if (first.isMissing(i) || second.isMissing(i)){ if (first.isMissing(i) && second.isMissing(i)) { diff = 1; } else { if (second.isMissing(i)) { diff = norm(first.value(i), i); } else { diff = norm(second.value(i), i); } if (diff < 0.5) { diff = 1.0 - diff; } } } else { diff = norm(first.value(i), i) - norm(second.value(i), i); } distance += diff * diff; } } return distance; }
对每一个属性遍历,计算数值属性距离的平方和,norm方法为规范化距离公式,为【0,1】的实数
6)跟踪norm规范化方法,源码如下:
private double norm(double x,int i) { if (Double.isNaN(m_MinArray[i]) || Utils.eq(m_MaxArray[i], m_MinArray[i])) { return 0; } else { return (x - m_MinArray[i]) / (m_MaxArray[i] - m_MinArray[i]); } }
规范化距离:(x - m_MinArray[i]) / (m_MaxArray[i] - m_MinArray[i]);
具体的算法伪代码,请查找最近邻分类器的论文,我就不贴出来了。