接着上篇blog,继续看log里面的信息如下:
+ echo 'Training Naive Bayes model'
Training Naive Bayes model
+ ./bin/mahout trainnb -i /home/mahout/mahout-work-mahout/20news-train-vectors -el -o /home/mahout/mahout-work-mahout/model -li /home/mahout/mahout-work-mahout/labelindex -ow
这里mahout的trainnb对应的源码文件是TrainNaiveBayesJob类,该类主要的工作是:(1)新建了一个LabelIndex的文件;(2)执行了一个prepareJob,Mapper和Reducer分别是:IndexInstancesMapper、VectorSumReducer;(2)执行了另外的一个prepareJob,Mapper和Reducer分别是:WeightsMapper、VectorSumReducer;本篇主要分析前面两个工作。
新建LabelIndex的代码如下:
private long createLabelIndex(Path labPath) throws IOException {long labelSize = 0;if (hasOption(LABELS)) {Iterable labels = Splitter.on(",").split(getOption(LABELS));labelSize = BayesUtils.writeLabelIndex(getConf(), labels, labPath);} else if (hasOption(EXTRACT_LABELS)) {SequenceFileDirIterable iterable =new SequenceFileDirIterable(getInputPath(), PathType.LIST, PathFilters.logsCRCFilter(), getConf());labelSize = BayesUtils.writeLabelIndex(getConf(), labPath, iterable);}return labelSize;}
这里的主要工作是把相关的文件名转换为数字,文件名如下图:
下面看Mapper,IndexInstancesMapper的主要代码如下:
labelIndex = BayesUtils.readIndexFromCache(ctx.getConfiguration());
String label = labelText.toString().split("/")[1]; if (labelIndex.containsKey(label)) {ctx.write(new IntWritable(labelIndex.get(label)), instance);
首先在setup函数中读取labelindex的map映射关系,然后在map中针对输入/alt.atheism/51060解析/后面的字符串,即文件名进行匹配,输出对应的数字和相应的value不变;
VectorSumReducer:
Vector vector = null;for (VectorWritable v : values) {if (vector == null) {vector = v.get();} else {vector.assign(v.get(), Functions.PLUS);}}ctx.write(key, new VectorWritable(vector));
上面的代码就是把相同的文件对应的word的单词的个数全部加起来,由于一共有20个文件,所以这里的reduce输出应该有20个,对应log里面的信息,可以看到确实匹配,如下图:
这里额可以通过下面的代码来测试相关的文件:
package mahout.fansy.test.bayes.read;import java.io.IOException;
import java.net.URI;
import java.util.HashMap;
import java.util.Map;import org.apache.hadoop.conf.Configuration;
import org.apache.hadoop.fs.FileSystem;
import org.apache.hadoop.fs.Path;
import org.apache.hadoop.io.IOUtils;
import org.apache.hadoop.io.IntWritable;
import org.apache.hadoop.io.SequenceFile;
import org.apache.hadoop.io.Text;
import org.apache.hadoop.io.Writable;
import org.apache.hadoop.util.ReflectionUtils;
import org.apache.mahout.math.VectorWritable;public class ReadLabelIndex {/*** @param args*/public static Configuration conf=new Configuration();static String fPath="";static String trainPath="";static{conf.set("mapred.job.tracker", "ubuntu:9001");fPath="hdfs://ubuntu:9000/home/mahout/mahout-work-mahout/labelindex"; // lableindex 数据文件trainPath="hdfs://ubuntu:9000/home/mahout/mahout-work-mahout/"+"20news-train-vectors/part-r-00000"; // 训练样本数据}public static void main(String[] args) throws IOException {// readFromFile(fPath);readFromFile(trainPath);}/*** 读取LabelIndex文件* @param fPath* @return* @throws IOException*/public static Map readFromFile(String fPath) throws IOException{FileSystem fs = FileSystem.get(URI.create(fPath), conf);Path path = new Path(fPath);Map map=new HashMap();SequenceFile.Reader reader = null;try {reader = new SequenceFile.Reader(fs, path, conf);Writable key = (Writable)ReflectionUtils.newInstance(reader.getKeyClass(), conf);Writable value = (Writable)ReflectionUtils.newInstance(reader.getValueClass(), conf);while (reader.next(key, value)) {// Writable k=; // 如何实现Writable的深度复制?// map.put(key, value);System.out.println(key.toString()+", "+value.toString());System.exit(-1);// 只打印第一条记录}} finally {IOUtils.closeStream(reader);}return map;}}
这里在写的时候想做一个通用的,所以需要对Writable深度复制,但是一时间还没有想到办法,所以这里留个问题,有时间解决。
分享,成长,快乐
转载请注明blog地址:http://blog.csdn.net/fansy1990