WekaVoting源代码分析.docVIP

Weka[26] Voting 源代码分析 作者：Koala++/屈伟 Voting 这个类和 Bagging 差不多，不过我感觉 Voting 更好用一点，Bagging 继承关系看 起 来 可 能 有 很 强 的 扩 展 性 ， 但 对 于 我 ， 这 一 点 没 什 么 意 义 。 Voting 继 承 自 RandomizableMultipleClassifiersCombiner，而它又继承自 MultipleClassifiersCombiner，它继承 自 Classifier。 下面看一下 Voting 的成员变量： /** combination rule: Average of Probabilities */ public static final int AVERAGE_RULE = 1; /** combination rule: Product of Probabilities (only nominal classes) */ public static final int PRODUCT_RULE = 2; /** combination rule: Majority Voting (only nominal classes) */ public static final int MAJORITY_VOTING_RULE = 3; /** combination rule: Minimum Probability */ public static final int MIN_RULE = 4; /** combination rule: Maximum Probability */ public static final int MAX_RULE = 5; /** combination rule: Median Probability (only numeric class) */ public static final int MEDIAN_RULE = 6; /** combination rules */ public static final Tag[] TAGS_RULES = { new Tag(AVERAGE_RULE, AVG, Average of Probabilities), new Tag(PRODUCT_RULE, PROD, Product of Probabilities), new Tag(MAJORITY_VOTING_RULE, MAJ, Majority Voting), new Tag(MIN_RULE, MIN, Minimum Probability), new Tag(MAX_RULE, MAX, Maximum Probability), new Tag(MEDIAN_RULE, MED, Median) }; /** Combination Rule variable */ protected int m_CombinationRule = AVERAGE_RULE; 比起 Bagging 是丰富了不少，另外，源代码上列出来参考文献我没找到，事实上也没必 要，直接看代码就可以了。 public void buildClassifier(Instances data) throws Exception { // can classifier handle the data? getCapabilities().testWithFail(data); // remove instances with missing class Instances newData = new Instances(data); newData.deleteWithMissingClass(); m_Random = new Random(getSeed()); for (int i = 0; i m_Classifiers.length; i++) { getClassifier(i).buildClassifier(newData); } } 很简单，就是训练多个分类器。 下面是 distributionForInstance 的代码： public double[] distributionForInstance(Instance instance) throws Exception { double[] result = new double[instance.numClasses()]; switch (m_CombinationRule) { case AVERAGE_RULE: result = dist