private double entropy(Map<String, String> specifiedAttributes) {
double totalExamples = records.count();
double positiveExamples = records.countPositive(specifiedAttributes);
double negativeExamples = records.countNegative(specifiedAttributes);
return -nlog2(positiveExamples / totalExamples) - nlog2(negativeExamples / totalExamples);
}
private double entropy(
String attribute, String decision, Map<String, String> specifiedAttributes) {
double totalExamples = records.count(attribute, decision, specifiedAttributes);
double positiveExamples = records.countPositive(attribute, decision, specifiedAttributes);
double negativeExamples = records.countNegative(attribute, decision, specifiedAttributes);
// logger.info("positiveExamples is --> {}.", positiveExamples);
// logger.info("negativeExamples is --> {}.", negativeExamples);
// logger.info("totalExamples is --> {}.", totalExamples);
if (positiveExamples == 0 || negativeExamples == 0 || totalExamples == 0) return 0;
return -nlog2(positiveExamples / totalExamples) - nlog2(negativeExamples / totalExamples);
}
private double informationGain(String attribute, Map<String, String> specifiedAttributes) {
double sum = entropy(specifiedAttributes);
double examplesCount = records.count(specifiedAttributes);
if (examplesCount == 0) return sum;
Map<String, Set<String>> decisions = records.extractDecisions();
for (String decision : decisions.get(attribute)) {
double entropyPart = entropy(attribute, decision, specifiedAttributes);
// logger.info("entropyPart is --> {}.", entropyPart);
double decisionCount = records.countDecisions(attribute, decision);
sum += -(decisionCount / examplesCount) * entropyPart;
}
return sum;
}
