KNN算法又称为k近邻分类(k-nearest neighbor classification)算法。最简单平凡的分类器。KNN算法则是从训练集中找到和新数据最接近的k条记录,然后根据他们的主要分类来决定新数据的类别。该算法涉及3个主要因素:训练集、距离、k的大小。可用于客户流失预测、欺诈侦测等(更适合于稀有事件的分类问题)。
计算步骤如下:
1)计算距离:给定测试对象,计算它与训练集中的每个对象的距离。距离计算可以选择欧氏距离、曼哈顿距离、余弦距离等。计算距离之前最好对数据进行规范化处理,以便于更好的计算。
2)寻找邻居:圈定距离最近的k个训练对象,作为测试对象的近邻。K值的选择可以通过若干试验,选取分类误差最小的K值。
3)判断分类:根据这k个近邻归属的主要类别,来对测试对象分类。判定方式主要是投票决定,少数服从多数,近邻中哪个类别的点最多就分为该类。也可以通过加权投票方法来决定。
优点
简单,易于理解,易于实现,无需估计参数,无需训练
适合对稀有事件进行分类(例如当流失率很低时,比如低于0.5%,构造流失预测模型)
特别适合于多分类问题(multi-modal,对象具有多个类别标签),例如根据基因特征来判断其功能分类,KNN比SVM的表现要好
缺点
计算开销大,需要有效的存储技术和并行硬件的支撑。
可解释性较差,无法给出决策树那样的规则。
下面是基于MapReduce简单实现KNN的代码:
public class KNNClassifier {
private static void configureJob(Job job) {
job.setJarByClass(KNNClassifier.class);
job.setMapperClass(KNNMapper.class);
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(PointWritable.class);
job.setReducerClass(KNNReducer.class);
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);
job.setInputFormatClass(TextInputFormat.class);
job.setOutputFormatClass(TextOutputFormat.class);
}
public static void main(String[] args) {
long start = System.currentTimeMillis();
Configuration configuration = new Configuration();
try {
String[] inputArgs = new GenericOptionsParser(
configuration, args).getRemainingArgs();
if (inputArgs.length != 4) {
System.out.println("error, please input three path.");
System.out.println("1 train set path.");
System.out.println("2 test set path.");
System.out.println("3 output path.");
System.out.println("4 k value.");
System.exit(2);
}
DistributedCache.addCacheFile(new Path(inputArgs[0]).toUri(), configuration);
configuration.set("k", inputArgs[3]);
Job job = new Job(configuration, "KNN Classifier");
FileInputFormat.setInputPaths(job, new Path(inputArgs[1]));
FileOutputFormat.setOutputPath(job, new Path(inputArgs[2]));
configureJob(job);
System.out.println(job.waitForCompletion(true) ? 0 : 1);
long end = System.currentTimeMillis();
System.out.println("spend time: " + (end - start) / 1000);
} catch (Exception e) {
e.printStackTrace();
}
}
}
class KNNMapper extends Mapper<LongWritable, Text, Text, PointWritable> {
private List<Point> trainPoints = new ArrayList<Point>();
@Override
protected void setup(Context context) throws IOException, InterruptedException {
super.setup(context);
Configuration conf = context.getConfiguration();
FileSystem fs = FileSystem.get(conf);
URI[] uris = DistributedCache.getCacheFiles(conf);
Path[] paths = HDFSUtils.getPathFiles(fs, new Path(uris[0]));
for(Path path : paths) {
FSDataInputStream in = fs.open(path);
BufferedReader reader = new BufferedReader(new InputStreamReader(in));
String line = reader.readLine();
while (null != line && !"".equals(line)) {
String[] datas = line.split(" ");
trainPoints.add(new Point(Double.parseDouble(datas[0]),
Double.parseDouble(datas[1]), datas[2]));
line = reader.readLine();
}
IOUtils.closeQuietly(in);
IOUtils.closeQuietly(reader);
}
}
@Override
protected void map(LongWritable key, Text value, Context context)
throws IOException, InterruptedException {
String line = value.toString();
String[] datas = line.split(" ");
double x = Double.parseDouble(datas[0]);
double y = Double.parseDouble(datas[1]);
Point testPoint = new Point(x, y);
String outputKey = x + "-" + y;
for (Point trainPoint : trainPoints) {
double distance = distance(testPoint, trainPoint);
context.write(new Text(outputKey), new PointWritable(trainPoint, distance));
}
}
public double distance(Point point1, Point point2) {
return Math.sqrt(Math.pow((point1.getX() - point2.getX()), 2)
+ Math.pow((point1.getY() - point2.getY()), 2));
}
@Override
protected void cleanup(Context context) throws IOException, InterruptedException {
super.cleanup(context);
}
}
class KNNReducer extends Reducer<Text, PointWritable, Text, Text> {
private int k = 0;
@Override
protected void setup(Context context) throws IOException, InterruptedException {
super.setup(context);
Configuration conf = context.getConfiguration();
k = Integer.parseInt(conf.get("k", "0"));
}
@Override
protected void reduce(Text key, Iterable<PointWritable> values,
Context context) throws IOException, InterruptedException {
System.out.println(key);
List<PointWritable> points = new ArrayList<PointWritable>();
for (PointWritable point : values) {
points.add(new PointWritable(point));
}
Collections.sort(points, new Comparator<PointWritable>() {
@Override
public int compare(PointWritable o1, PointWritable o2) {
return o1.getDistance().compareTo(o2.getDistance());
}
});
Map<String, Integer> map = new HashMap<String, Integer>();
k = points.size() < k ? points.size() : k;
for (int i = 0; i < k; i++) {
PointWritable point = points.get(i);
String category = point.getCategory().toString();
Integer count = map.get(category);
map.put(category, null == count ? 1 : count + 1);
}
List<Map.Entry<String, Integer>> list =
new ArrayList<Map.Entry<String, Integer>>(map.entrySet());
Collections.sort(list, new Comparator<Map.Entry<String, Integer>>(){
@Override
public int compare(Entry<String, Integer> o1,
Entry<String, Integer> o2) {
return o2.getValue().compareTo(o1.getValue());
}
});
context.write(key, new Text(list.get(0).getKey()));
}
@Override
protected void cleanup(Context context) throws IOException,
InterruptedException {
super.cleanup(context);
}
}
代码托管:https://github.com/fighting-one-piece/repository-datamining.git