Naive Bayes在mapreduce上的实现

Naive Bayes是比较常用的分类器，因为思想比较简单。之所以说是naive，是因为他假设用于分类的特征在类确定的条件下是条件独立的，这个假设使得分类变得很简单，但会损失一定的精度。
具体推导可以看《统计学习方法》
经过推导我们可知y=argMaxP(Y=ck)*P(X=x|Y=ck)。那么我们需要求先验概率也就是P（Y=ck）和求条件概率p(X=x|Y=ck).
具体的例子以：http://blog.163.com/jiayouweijiewj@126/blog/static/1712321772010102802635243/来说明。
我这里一共用了4个mapreduce，因为采用了多项式模型，先验概率P(c)= 类c下单词总数/整个训练样本的单词总数。类条件概率P(tk|c)=(类c下单词tk在各个文档中出现过的次数之和+1)/(类c下单词总数+|V|)（|V|是单词种类数）。输入是：
1:Chinese Beijing Chinese
1:Chinese Chinese Shanghai
1:Chinese Macao
0:Tokyo Japan Chinese
1 一个mapreduce是用于求在各个类别下的单词数，这个是为了后面求先验概率用的。
输出为：
0 3
1 8

2 一个mapreduce用于求条件概率，输出为：
0:Chinese 0.2222222222222222
0:Japan 0.2222222222222222
0:Tokyo 0.2222222222222222
1:Beijing 0.14285714285714285
1:Chinese 0.42857142857142855
1:Macao 0.14285714285714285
1:Shanghai 0.14285714285714285

3 一个mapreduce用于计算单词种类数，输出为：
num is 6
4 最后一个mapreduce是用于预测的。

下面说下各个mapreduce的实现：
1 求各个类别下的单词数，这个比较简单，就是以类别为key，然后进行单词统计就好。
附上代码：

package hadoop.MachineLearning.Bayes.Pro;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;

public class PriorProbability {//用于求各个类别下的单词数，为后面求先验概率

public static void main(String[] args) throws Exception {
Configuration conf = new Configuration();
String input="hdfs://10.107.8.110:9000/Bayes/Bayes_input/";
String output="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Pro/";
Job job = Job.getInstance(conf, "ProirProbability");
job.setJarByClass(hadoop.MachineLearning.Bayes.Pro.PriorProbability.class);
// TODO: specify a mapper
job.setMapperClass(MyMapper.class);
//job.setMapInputKeyClass(LongWritable.class);
// TODO: specify a reducer
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(Text.class);
job.setReducerClass(MyReducer.class);

// TODO: specify output types
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);

// TODO: specify input and output DIRECTORIES (not files)
FileInputFormat.setInputPaths(job, new Path(input));
FileOutputFormat.setOutputPath(job, new Path(output));

if (!job.waitForCompletion(true))
return;
}

}

package hadoop.MachineLearning.Bayes.Pro;

import java.io.IOException;

import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;
import org.apache.hadoop.mapreduce.Mapper.Context;

public class MyMapper extends Mapper<LongWritable, Text, Text, Text> {

public void map(LongWritable ikey, Text ivalue, Context context)
throws IOException, InterruptedException {
String[] line=ivalue.toString().split(":| ");
int size=line.length-1;
context.write(new Text(line[0]),new Text(String.valueOf(size)));
}

}

package hadoop.MachineLearning.Bayes.Pro;

import java.io.IOException;

import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;
import org.apache.hadoop.mapreduce.Reducer.Context;

public class MyReducer extends Reducer<Text, Text, Text, IntWritable> {

public void reduce(Text _key, Iterable<Text> values, Context context)
throws IOException, InterruptedException {
// process values
int sum=0;
for (Text val : values) {
sum+=Integer.parseInt(val.toString());
}
context.write(_key,new IntWritable(sum));
}

}

2 求文档中的单词种类数，自己实现的方法不太好，思路是，对每一行的输入都以相同的key输出，然后在combiner中先利用set求得该节点上的不重复的单词，接着在reduce中再利用set，将所有单词求种类数。感觉好一点的话是先按照单词进行规约，最后再利用一个mapreduce对单词种类数进行统计。但是考虑到刚学会mapreduce不久还不会写链式，而且一个bayes已经写了4个mapreduce就不考虑再复杂化了。

package hadoop.MachineLearning.Bayes.Count;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;

public class Count {//计算文档中的单词种类数目

public static void main(String[] args) throws Exception {
Configuration conf = new Configuration();
Job job = Job.getInstance(conf, "Count");
String input="hdfs://10.107.8.110:9000/Bayes/Bayes_input";
String output="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Count";
job.setJarByClass(hadoop.MachineLearning.Bayes.Count.Count.class);
// TODO: specify a mapper
job.setMapperClass(MyMapper.class);
// TODO: specify a reducer
job.setCombinerClass(MyCombiner.class);
job.setReducerClass(MyReducer.class);

// TODO: specify output types
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(Text.class);

// TODO: specify input and output DIRECTORIES (not files)
FileInputFormat.setInputPaths(job, new Path(input));
FileOutputFormat.setOutputPath(job, new Path(output));

if (!job.waitForCompletion(true))
return;
}

}

package hadoop.MachineLearning.Bayes.Count;

import java.io.IOException;

import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;

public class MyMapper extends Mapper<LongWritable, Text, Text, Text> {

public void map(LongWritable ikey, Text ivalue, Context context)
throws IOException, InterruptedException {
String[] line=ivalue.toString().split(":| ");
String key="1";
System.out.println("   ");
System.out.println("   ");
System.out.println("   ");
for(int i=1;i<line.length;i++){

System.out.println(line[i]);
context.write(new Text(key),new Text(line[i]));//以相同的key进行输出，使得能最后输出到一个reduce中
}
}

}

package hadoop.MachineLearning.Bayes.Count;

import java.io.IOException;
import java.util.HashSet;
import java.util.Iterator;
import java.util.Set;

import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class MyCombiner extends Reducer<Text, Text, Text, Text> {//先在本地的节点上利用set删去重复的单词

public void reduce(Text _key, Iterable<Text> values, Context context)
throws IOException, InterruptedException {
// process values
Set set=new HashSet();
for (Text val : values) {
set.add(val.toString());
}
for(Iterator it=set.iterator();it.hasNext();){
context.write(new Text("1"),new Text(it.next().toString()));
}
}

}

package hadoop.MachineLearning.Bayes.Count;

import java.io.IOException;
import java.util.HashSet;
import java.util.Set;

import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class MyReducer extends Reducer<Text, Text, Text, Text> {//通过combiner后，再利用set对单词进行去重，最后得到种类数

public void reduce(Text _key, Iterable<Text> values, Context context)
throws IOException, InterruptedException {
// process values
Set set=new HashSet();
for (Text val : values) {
set.add(val.toString());
}
context.write(new Text("num is "),new Text(String.valueOf(set.size())));
}

}

3 求条件概率.这里需要用到该类别下该单词的数目sum,该类别下的单词总数，文档中的单词种类数。这些都可以在之前的输出文件中获得，我这里都用map去接受这些数据。由于有些单词没有出现在该类别下，例如P(Japan | yes)=P(Tokyo | yes)，如果将他们当作0处理，那么导致该条件概率会是0，所以这里用了平滑的方法可以参考上述的链接。这里有个细节，就是条件概率生成的会比较多，需要一种高效的存储和查找方式，我这里因为水平不够，就直接用map来存放了，如果对于大的数据，这个会很低效。

package hadoop.MachineLearning.Bayes.Cond;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;

public class CondiPro {//用于求条件概率

public static void main(String[] args) throws Exception {
Configuration conf = new Configuration();
String input="hdfs://10.107.8.110:9000/Bayes/Bayes_input";
String output="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Con";
String proPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Pro";//这是之前求各个类别下单词数目的输出
String countPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Count";//这是之前求的单词种类数
conf.set("propath",proPath);
conf.set("countPath",countPath);
Job job = Job.getInstance(conf, "ConditionPro");

job.setJarByClass(hadoop.MachineLearning.Bayes.Cond.CondiPro.class);
// TODO: specify a mapper
job.setMapperClass(MyMapper.class);
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(IntWritable.class);
// TODO: specify a reducer
job.setReducerClass(MyReducer.class);

// TODO: specify output types
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(IntWritable.class);

// TODO: specify input and output DIRECTORIES (not files)
FileInputFormat.setInputPaths(job, new Path(input));
FileOutputFormat.setOutputPath(job, new Path(output));

if (!job.waitForCompletion(true))
return;
}

}

package hadoop.MachineLearning.Bayes.Cond;

import java.io.IOException;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;

public class MyMapper extends Mapper<LongWritable, Text, Text, IntWritable> {

public void map(LongWritable ikey, Text ivalue, Context context)
throws IOException, InterruptedException {
String[] line=ivalue.toString().split(":| ");
for(int i=1;i<line.length;i++){
String key=line[0]+":"+line[i];
context.write(new Text(key),new IntWritable(1));
}
}

}

package hadoop.MachineLearning.Bayes.Cond;

import java.io.IOException;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class MyReducer extends Reducer<Text, IntWritable, Text, DoubleWritable> {
public Map<String,Integer> map;
public int count=0;
public void setup(Context context) throws IOException{
Configuration conf=context.getConfiguration();

String proPath=conf.get("propath");
String countPath=conf.get("countPath");//
map=Utils.getMapFormHDFS(proPath);//获得各个类别下的单词数
count=Utils.getCountFromHDFS(countPath);//获得单词种类数
}
public void reduce(Text _key, Iterable<IntWritable> values, Context context)
throws IOException, InterruptedException {
// process values
int sum=0;
for (IntWritable val : values) {
sum+=val.get();
}
int type=Integer.parseInt(_key.toString().split(":")[0]);
double probability=0.0;
for(Map.Entry<String,Integer> entry:map.entrySet()){
if(type==Integer.parseInt(entry.getKey())){
probability=(sum+1)*1.0/(entry.getValue()+count);//条件概率的计算
}
}
context.write(_key,new DoubleWritable(probability));
}

}

package hadoop.MachineLearning.Bayes.Cond;

import java.io.IOException;
import java.util.HashMap;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.util.LineReader;

public class Utils {

/**
* @param args
* @throws IOException
*/

public static Map<String,Integer> getMapFormHDFS(String input) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);
Map<String,Integer> map=new HashMap();
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
map.put(temp[0],Integer.parseInt(temp[1]));
}
reader.close();
}
}

return map;

}

public static Map<String,Double> getMapFormHDFS(String input,boolean j) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);
Map<String,Double> map=new HashMap();
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
map.put(temp[0],Double.parseDouble(temp[1]));
}
reader.close();
}
}

return map;

}

public static int getCountFromHDFS(String input) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);

int count=0;
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
count=Integer.parseInt(temp[1]);
}
reader.close();
}
}
return count;
}

public static void main(String[] args) throws IOException {
// TODO Auto-generated method stub
String proPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Pro";
String countPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Count/";
Map<String,Integer> map=Utils.getMapFormHDFS(proPath);
for(Map.Entry<String,Integer> entry:map.entrySet()){
System.out.println(entry.getKey()+"->"+entry.getValue());
}

int count=Utils.getCountFromHDFS(countPath);
System.out.println("count is "+count);
}

}

4 预测，例如输入Chinese, Chinese, Chinese, Tokyo, Japan。那就分别对每个单词以0，1的类别进行输出，输出为type:words，接着就是在条件概率中查找，进行简单的累乘即可。

package hadoop.MachineLearning.Bayes.Predict;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Job;
import org.apache.hadoop.mapreduce.lib.input.FileInputFormat;
import org.apache.hadoop.mapreduce.lib.output.FileOutputFormat;

public class Predict {

public static void main(String[] args) throws Exception {//预测
Configuration conf = new Configuration();
String input="hdfs://10.107.8.110:9000/Bayes/Predict_input";
String output="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Predict";
String condiProPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Con";
String proPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Pro";
String countPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Count";
conf.set("condiProPath",condiProPath);
conf.set("proPath",proPath);
conf.set("countPath",countPath);
Job job = Job.getInstance(conf, "Predict");
job.setJarByClass(hadoop.MachineLearning.Bayes.Predict.Predict.class);
// TODO: specify a mapper
job.setMapperClass(MyMapper.class);
job.setMapOutputKeyClass(Text.class);
job.setMapOutputValueClass(Text.class);
// TODO: specify a reducer
job.setReducerClass(MyReducer.class);

// TODO: specify output types
job.setOutputKeyClass(Text.class);
job.setOutputValueClass(DoubleWritable.class);

// TODO: specify input and output DIRECTORIES (not files)
FileInputFormat.setInputPaths(job, new Path(input));
FileOutputFormat.setOutputPath(job, new Path(output));

if (!job.waitForCompletion(true))
return;
}

}

package hadoop.MachineLearning.Bayes.Predict;

import java.io.IOException;
import java.util.HashMap;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.LongWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Mapper;

public class MyMapper extends Mapper<LongWritable, Text, Text, Text> {

public Map<String,Integer> map=new HashMap();

public void setup(Context context) throws IOException{
Configuration conf=context.getConfiguration();
String proPath=conf.get("proPath");
map=Utils.getMapFormHDFS(proPath);
}

public void map(LongWritable ikey, Text ivalue, Context context)
throws IOException, InterruptedException {
for(Map.Entry<String,Integer> entry:map.entrySet()){
context.write(new Text(entry.getKey()),ivalue);//对每一行数据，打上所有类别，方便后续的求条件概率
}
}

}

package hadoop.MachineLearning.Bayes.Predict;

import java.io.IOException;
import java.util.HashMap;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.io.DoubleWritable;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.mapreduce.Reducer;

public class MyReducer extends Reducer<Text, Text, Text, DoubleWritable> {

public Map<String,Double> mapDouble=new HashMap();//存放条件概率

public Map<String,Integer> mapInteger=new HashMap();//存放各个类别下的单词数

public Map<String,Double> noFind=new HashMap();//用于那些单词没有出现在某个类别中的

public Map<String,Double> prePro=new HashMap();//求的后的先验概率

public void setup(Context context) throws IOException{
Configuration conf=context.getConfiguration();

String condiProPath=conf.get("condiProPath");
String proPath=conf.get("proPath");
String countPath=conf.get("countPath");
mapDouble=Utils.getMapFormHDFS(condiProPath,true);
mapInteger=Utils.getMapFormHDFS(proPath);
int count=Utils.getCountFromHDFS(countPath);
for(Map.Entry<String,Integer> entry:mapInteger.entrySet()){
double pro=0.0;
noFind.put(entry.getKey(),(1.0/(count+entry.getValue())));
}
int sum=0;
for(Map.Entry<String,Integer> entry:mapInteger.entrySet()){
sum+=entry.getValue();
}

for(Map.Entry<String,Integer> entry:mapInteger.entrySet()){
prePro.put(entry.getKey(),(entry.getValue()*1.0/sum));
}

}

public void reduce(Text _key, Iterable<Text> values, Context context)
throws IOException, InterruptedException {
// process values
String type=_key.toString();
double pro=1.0;
for (Text val : values) {
String[] words=val.toString().split(" ");
for(int i=0;i<words.length;i++){
String condi=type+":"+words[i];
if(mapDouble.get(condi)!=null){//如果该单词出现在该类别中，说明有条件概率
pro=pro*mapDouble.get(condi);
}else{//如果该单词不在该类别中，就采用默认的条件概率
pro=pro*noFind.get(type);
}
}
}
pro=pro*prePro.get(type);
context.write(new Text(type),new DoubleWritable(pro));
}

}

package hadoop.MachineLearning.Bayes.Predict;

import java.io.IOException;
import java.util.HashMap;
import java.util.Map;

import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.fs.FileStatus;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.util.LineReader;

public class Utils {

/**
* @param args
* @throws IOException
*/

public static Map<String,Integer> getMapFormHDFS(String input) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);
Map<String,Integer> map=new HashMap();
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
map.put(temp[0],Integer.parseInt(temp[1]));
}
reader.close();
}
}

return map;

}

public static Map<String,Double> getMapFormHDFS(String input,boolean j) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);
Map<String,Double> map=new HashMap();
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
map.put(temp[0],Double.parseDouble(temp[1]));
}
reader.close();
}
}

return map;

}

public static int getCountFromHDFS(String input) throws IOException{
Configuration conf=new Configuration();
Path path=new Path(input);
FileSystem fs=path.getFileSystem(conf);

FileStatus[] stats=fs.listStatus(path);

int count=0;
for(int i=0;i<stats.length;i++){
if(stats[i].isFile()){
FSDataInputStream infs=fs.open(stats[i].getPath());
LineReader reader=new LineReader(infs,conf);
Text line=new Text();
while(reader.readLine(line)>0){
String[] temp=line.toString().split("    ");
//System.out.println(temp.length);
count=Integer.parseInt(temp[1]);
}
reader.close();
}
}
return count;
}

public static void main(String[] args) throws IOException {
// TODO Auto-generated method stub
String proPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Pro";
String countPath="hdfs://10.107.8.110:9000/Bayes/Bayes_output/Count/";
Map<String,Integer> map=Utils.getMapFormHDFS(proPath);
for(Map.Entry<String,Integer> entry:map.entrySet()){
System.out.println(entry.getKey()+"->"+entry.getValue());
}

int count=Utils.getCountFromHDFS(countPath);
System.out.println("count is "+count);
}

}