基于hadoop的推荐算法-mahout版
基于hadoop的推荐算法，讲其中mahout实现的基于项目的推荐算法分为4步：
1.获得人-物 用户矩阵
输入为所有人对物品的评价或关联
map端输出key为人，value为物品+倾好度
reeduce端输出key为人，vallue为多个物品+倾好度
2.获得物-物 项目矩阵
输入为“用户矩阵”，讲每一行人-物数据中的物品做笛卡尔积，生产成物-物的关联
map端输出为key为物，value为关联度
reduce端输出key为物，value为多个物的关联度
（可以根据各种规则生成项目相似度矩阵表，此处算法带过）
修改：
求项目相似矩阵是基于项目的协同过滤算法的核心
公式有很多种，核心是物品i和物品j相关用户的交集与并集的商
mahout使用的公式是1.dot(i,j) = sum(Pi(u)*Pi(u))
2.norms(i) = sum(Pi(u)^2)
3.simi(i,j) = 1/(1+(norms(i)-2*dot(i,j)+noorm(i))^1/2)
mahout的实现方法是
第一个job，用物品-人的矩阵，求得norms，即物品的用户平方和，输出是物-norms
第二个job，Map：用人-物的矩阵，求Pi(u)*Pi(u),即相同用户的物品的评价的乘机，输出物-多个对端物品的Pi(u)*Pi(u)
Reduce：用物-多个对端物品的Pi(u)*Pi(u)和物-norms，求得物品的相似矩阵（因为这个时候可以汇总所有和这个物品相关的物品的dot）
第三个job，补全物品的相似矩阵
3.获得用户-项目相似矩阵
输入为人-物 用户矩阵 和 物-物 项目矩阵
Map端输出key为物，value为类VectorOrPrefWritable，是包含物与人的倾好度，或是物与物的相似度
reduce端输出key为物，value为类VectorAndPrefWritable,是汇总当个物品到所有人的倾好度和到所有物品的相似度
4.获得用户推荐矩阵
输入为VectorAndPrefWritable
Map端输出为key：人，value：物+系数（map端根据单个物品贡献的系数生成推荐系数，也就是人到物品A的倾好度*物品A到其他物品的相似度）
reduce端输出为key：人，，value：推荐项目+系数（reduce端使用自定公式，汇总所有单物品贡献的四叔，求人到其他项目的倾好度，取topn作为当前用户的推荐项目）
再在这里贴几个mahout推荐算法分析的帖子：
http://eric-gcm.iyunv.com/blog/1817822
http://eric-gcm.iyunv.com/blog/1818033
http://eric-gcm.iyunv.com/blog/1820060
以下是mahout代码：
ItemSimilarityJob类是mahout使用hadoop做推荐引擎的主要实现类，下面开始分析。
run()函数是启动函数：
Java代码
[*]public final class RecommenderJob extends AbstractJob {
[*]
[*] public static final String BOOLEAN_DATA = "booleanData";
[*]
[*] private static final int DEFAULT_MAX_SIMILARITIES_PER_ITEM = 100;
[*] private static final int DEFAULT_MAX_PREFS_PER_USER = 1000;
[*] private static final int DEFAULT_MIN_PREFS_PER_USER = 1;
[*]
[*] @Override
[*] public int run(String[] args) throws Exception {
[*] //这里原来有大一堆代码，都是用来载入配置项，不用管它
[*]
[*] //第一步：准备矩阵，将原始数据转换为一个矩阵，在PreparePreferenceMatrixJob这个类中完成
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] ToolRunner.run(getConf(), new PreparePreferenceMatrixJob(), new String[]{
[*] "--input", getInputPath().toString(),
[*] "--output", prepPath.toString(),
[*] "--maxPrefsPerUser", String.valueOf(maxPrefsPerUserInItemSimilarity),
[*] "--minPrefsPerUser", String.valueOf(minPrefsPerUser),
[*] "--booleanData", String.valueOf(booleanData),
[*] "--tempDir", getTempPath().toString()});
[*]
[*] numberOfUsers = HadoopUtil.readInt(new Path(prepPath, PreparePreferenceMatrixJob.NUM_USERS), getConf());
[*] }
[*]
[*] //第二步：计算协同矩阵
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*]
[*] /* special behavior if phase 1 is skipped */
[*] if (numberOfUsers == -1) {
[*] numberOfUsers = (int) HadoopUtil.countRecords(new Path(prepPath, PreparePreferenceMatrixJob.USER_VECTORS),
[*] PathType.LIST, null, getConf());
[*] }
[*]
[*] /* Once DistributedRowMatrix uses the hadoop 0.20 API, we should refactor this call to something like
[*] * new DistributedRowMatrix(...).rowSimilarity(...) */
[*] //calculate the co-occurrence matrix
[*] ToolRunner.run(getConf(), new RowSimilarityJob(), new String[]{
[*] "--input", new Path(prepPath, PreparePreferenceMatrixJob.RATING_MATRIX).toString(),
[*] "--output", similarityMatrixPath.toString(),
[*] "--numberOfColumns", String.valueOf(numberOfUsers),
[*] "--similarityClassname", similarityClassname,
[*] "--maxSimilaritiesPerRow", String.valueOf(maxSimilaritiesPerItem),
[*] "--excludeSelfSimilarity", String.valueOf(Boolean.TRUE),
[*] "--threshold", String.valueOf(threshold),
[*] "--tempDir", getTempPath().toString()});
[*] }
[*]
[*] //start the multiplication of the co-occurrence matrix by the user vectors
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] Job prePartialMultiply1 = prepareJob(
[*] similarityMatrixPath, prePartialMultiplyPath1, SequenceFileInputFormat.class,
[*] SimilarityMatrixRowWrapperMapper.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] Reducer.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] SequenceFileOutputFormat.class);
[*] boolean succeeded = prePartialMultiply1.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] //continue the multiplication
[*] Job prePartialMultiply2 = prepareJob(new Path(prepPath, PreparePreferenceMatrixJob.USER_VECTORS),
[*] prePartialMultiplyPath2, SequenceFileInputFormat.class, UserVectorSplitterMapper.class, VarIntWritable.class,
[*] VectorOrPrefWritable.class, Reducer.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] SequenceFileOutputFormat.class);
[*] if (usersFile != null) {
[*] prePartialMultiply2.getConfiguration().set(UserVectorSplitterMapper.USERS_FILE, usersFile);
[*] }
[*] prePartialMultiply2.getConfiguration().setInt(UserVectorSplitterMapper.MAX_PREFS_PER_USER_CONSIDERED,
[*] maxPrefsPerUser);
[*] succeeded = prePartialMultiply2.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] //finish the job
[*] Job partialMultiply = prepareJob(
[*] new Path(prePartialMultiplyPath1 + "," + prePartialMultiplyPath2), partialMultiplyPath,
[*] SequenceFileInputFormat.class, Mapper.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] ToVectorAndPrefReducer.class, VarIntWritable.class, VectorAndPrefsWritable.class,
[*] SequenceFileOutputFormat.class);
[*] setS3SafeCombinedInputPath(partialMultiply, getTempPath(), prePartialMultiplyPath1, prePartialMultiplyPath2);
[*] succeeded = partialMultiply.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] }
[*]
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] //filter out any users we don't care about
[*] /* convert the user/item pairs to filter if a filterfile has been specified */
[*] if (filterFile != null) {
[*] Job itemFiltering = prepareJob(new Path(filterFile), explicitFilterPath, TextInputFormat.class,
[*] ItemFilterMapper.class, VarLongWritable.class, VarLongWritable.class,
[*] ItemFilterAsVectorAndPrefsReducer.class, VarIntWritable.class, VectorAndPrefsWritable.class,
[*] SequenceFileOutputFormat.class);
[*] boolean succeeded = itemFiltering.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] }
[*]
[*] String aggregateAndRecommendInput = partialMultiplyPath.toString();
[*] if (filterFile != null) {
[*] aggregateAndRecommendInput += "," + explicitFilterPath;
[*] }
[*] //extract out the recommendations
[*] Job aggregateAndRecommend = prepareJob(
[*] new Path(aggregateAndRecommendInput), outputPath, SequenceFileInputFormat.class,
[*] PartialMultiplyMapper.class, VarLongWritable.class, PrefAndSimilarityColumnWritable.class,
[*] AggregateAndRecommendReducer.class, VarLongWritable.class, RecommendedItemsWritable.class,
[*] TextOutputFormat.class);
[*] Configuration aggregateAndRecommendConf = aggregateAndRecommend.getConfiguration();
[*] if (itemsFile != null) {
[*] aggregateAndRecommendConf.set(AggregateAndRecommendReducer.ITEMS_FILE, itemsFile);
[*] }
[*]
[*] if (filterFile != null) {
[*] setS3SafeCombinedInputPath(aggregateAndRecommend, getTempPath(), partialMultiplyPath, explicitFilterPath);
[*] }
[*] setIOSort(aggregateAndRecommend);
[*] aggregateAndRecommendConf.set(AggregateAndRecommendReducer.ITEMID_INDEX_PATH,
[*] new Path(prepPath, PreparePreferenceMatrixJob.ITEMID_INDEX).toString());
[*] aggregateAndRecommendConf.setInt(AggregateAndRecommendReducer.NUM_RECOMMENDATIONS, numRecommendations);
[*] aggregateAndRecommendConf.setBoolean(BOOLEAN_DATA, booleanData);
[*] boolean succeeded = aggregateAndRecommend.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] }
[*]
[*] return 0;
[*] }
第二步，计算协同矩阵，主要在RowSimilarityJob 这个类中完成
Java代码
[*]ToolRunner.run(getConf(), new RowSimilarityJob(), new String[]{
[*] "--input", new Path(prepPath, PreparePreferenceMatrixJob.RATING_MATRIX).toString(),
[*] "--output", similarityMatrixPath.toString(),
[*] "--numberOfColumns", String.valueOf(numberOfUsers),
[*] "--similarityClassname", similarityClassname,
[*] "--maxSimilaritiesPerRow", String.valueOf(maxSimilaritiesPerItem),
[*] "--excludeSelfSimilarity", String.valueOf(Boolean.TRUE),
[*] "--threshold", String.valueOf(threshold),
[*] "--tempDir", getTempPath().toString()});
[*] }
可以看到这个job的输入路径就是上一篇中，PreparePreferenceMatrixJob中最后一个reducer的输出路径。
下边详细分析RowSimilarityJob类的实现：
Java代码
[*]public class RowSimilarityJob extends AbstractJob {
[*]
[*]
[*] @Override
[*] public int run(String[] args) throws Exception {
[*] //一大堆载入参数的代码，忽略
[*]
[*] //第一个MapReduce
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] Job normsAndTranspose = prepareJob(getInputPath(), weightsPath, VectorNormMapper.class, IntWritable.class,
[*] VectorWritable.class, MergeVectorsReducer.class, IntWritable.class, VectorWritable.class);
[*] normsAndTranspose.setCombinerClass(MergeVectorsCombiner.class);
[*] Configuration normsAndTransposeConf = normsAndTranspose.getConfiguration();
[*] normsAndTransposeConf.set(THRESHOLD, String.valueOf(threshold));
[*] normsAndTransposeConf.set(NORMS_PATH, normsPath.toString());
[*] normsAndTransposeConf.set(NUM_NON_ZERO_ENTRIES_PATH, numNonZeroEntriesPath.toString());
[*] normsAndTransposeConf.set(MAXVALUES_PATH, maxValuesPath.toString());
[*] normsAndTransposeConf.set(SIMILARITY_CLASSNAME, similarityClassname);
[*] boolean succeeded = normsAndTranspose.waitForCompletion(true);
[*] if (!succeeded) {
[*] return -1;
[*] }
[*] }
[*] //第二个MapReduce
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] Job pairwiseSimilarity = prepareJob(weightsPath, pairwiseSimilarityPath, CooccurrencesMapper.class,
[*] IntWritable.class, VectorWritable.class, SimilarityReducer.class, IntWritable.class, VectorWritable.class);
[*] pairwiseSimilarity.setCombinerClass(VectorSumReducer.class);
[*] Configuration pairwiseConf = pairwiseSimilarity.getConfiguration();
[*] pairwiseConf.set(THRESHOLD, String.valueOf(threshold));
[*] pairwiseConf.set(NORMS_PATH, normsPath.toString());
[*] pairwiseConf.set(NUM_NON_ZERO_ENTRIES_PATH, numNonZeroEntriesPath.toString());
[*] pairwiseConf.set(MAXVALUES_PATH, maxValuesPath.toString());
[*] pairwiseConf.set(SIMILARITY_CLASSNAME, similarityClassname);
[*] pairwiseConf.setInt(NUMBER_OF_COLUMNS, numberOfColumns);
[*] pairwiseConf.setBoolean(EXCLUDE_SELF_SIMILARITY, excludeSelfSimilarity);
[*] boolean succeeded = pairwiseSimilarity.waitForCompletion(true);
[*] if (!succeeded) {
[*] return -1;
[*] }
[*] }
[*] //第三个MapReduce
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] Job asMatrix = prepareJob(pairwiseSimilarityPath, getOutputPath(), UnsymmetrifyMapper.class,
[*] IntWritable.class, VectorWritable.class, MergeToTopKSimilaritiesReducer.class, IntWritable.class,
[*] VectorWritable.class);
[*] asMatrix.setCombinerClass(MergeToTopKSimilaritiesReducer.class);
[*] asMatrix.getConfiguration().setInt(MAX_SIMILARITIES_PER_ROW, maxSimilaritiesPerRow);
[*] boolean succeeded = asMatrix.waitForCompletion(true);
[*] if (!succeeded) {
[*] return -1;
[*] }
[*] }
[*]
[*] return 0;
[*] }
可以看到RowSimilityJob也是分成三个MapReduce过程：
1、Mapper ：VectorNormMapper类，输出 ( userid_index, <itemid_index, pref> )类型
Java代码
[*]public static class VectorNormMapper extends Mapper<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] @Override
[*] protected void map(IntWritable row, VectorWritable vectorWritable, Context ctx)
[*] throws IOException, InterruptedException {
[*]
[*] Vector rowVector = similarity.normalize(vectorWritable.get());
[*]
[*] int numNonZeroEntries = 0;
[*] double maxValue = Double.MIN_VALUE;
[*]
[*] Iterator<Vector.Element> nonZeroElements = rowVector.iterateNonZero();
[*] while (nonZeroElements.hasNext()) {
[*] Vector.Element element = nonZeroElements.next();
[*] RandomAccessSparseVector partialColumnVector = new RandomAccessSparseVector(Integer.MAX_VALUE);
[*] partialColumnVector.setQuick(row.get(), element.get());
[*] //输出 ( userid_index, <itemid_index, pref> )类型
[*] ctx.write(new IntWritable(element.index()), new VectorWritable(partialColumnVector));
[*]
[*] numNonZeroEntries++;
[*] if (maxValue < element.get()) {
[*] maxValue = element.get();
[*] }
[*] }
[*]
[*] if (threshold != NO_THRESHOLD) {
[*] nonZeroEntries.setQuick(row.get(), numNonZeroEntries);
[*] maxValues.setQuick(row.get(), maxValue);
[*] }
[*] norms.setQuick(row.get(), similarity.norm(rowVector));
[*] //计算item的总数
[*] ctx.getCounter(Counters.ROWS).increment(1);
[*] }
[*]}
Reduer : MergeVectorsReducer类，输入的是(userid_index, <itemid_index, pref>)，同一个userid_index在此进行合并，输出（ userid_index, vector<itemid_index, pref> )
Java代码
[*] public static class MergeVectorsReducer extends Reducer<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] @Override
[*] protected void reduce(IntWritable row, Iterable<VectorWritable> partialVectors, Context ctx)
[*] throws IOException, InterruptedException {
[*] Vector partialVector = Vectors.merge(partialVectors);
[*]
[*] if (row.get() == NORM_VECTOR_MARKER) {
[*] Vectors.write(partialVector, normsPath, ctx.getConfiguration());
[*] } else if (row.get() == MAXVALUE_VECTOR_MARKER) {
[*] Vectors.write(partialVector, maxValuesPath, ctx.getConfiguration());
[*] } else if (row.get() == NUM_NON_ZERO_ENTRIES_VECTOR_MARKER) {
[*] Vectors.write(partialVector, numNonZeroEntriesPath, ctx.getConfiguration(), true);
[*] } else {
[*] ctx.write(row, new VectorWritable(partialVector));
[*] }
[*] }
[*] }
[*]}
2、Mapper：CooccurrencesMapper类，对同一个userid_index下的vector<itemid_index ,pref>进行处理，
收集<item1, item2>对， 输出为（ itemid_index, vector<itemid_index, value> ）
Java代码
[*]public static class CooccurrencesMapper extends Mapper<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] @Override
[*] protected void map(IntWritable column, VectorWritable occurrenceVector, Context ctx)
[*] throws IOException, InterruptedException {
[*] Vector.Element[] occurrences = Vectors.toArray(occurrenceVector);
[*] Arrays.sort(occurrences, BY_INDEX);
[*]
[*] int cooccurrences = 0;
[*] int prunedCooccurrences = 0;
[*] for (int n = 0; n < occurrences.length; n++) {
[*] Vector.Element occurrenceA = occurrences;
[*] Vector dots = new RandomAccessSparseVector(Integer.MAX_VALUE);
[*] for (int m = n; m < occurrences.length; m++) {
[*] Vector.Element occurrenceB = occurrences;
[*] if (threshold == NO_THRESHOLD || consider(occurrenceA, occurrenceB)) {
[*] dots.setQuick(occurrenceB.index(), similarity.aggregate(occurrenceA.get(), occurrenceB.get()));
[*] cooccurrences++;
[*] } else {
[*] prunedCooccurrences++;
[*] }
[*] }
[*] ctx.write(new IntWritable(occurrenceA.index()), new VectorWritable(dots));
[*] }
[*] ctx.getCounter(Counters.COOCCURRENCES).increment(cooccurrences);
[*] ctx.getCounter(Counters.PRUNED_COOCCURRENCES).increment(prunedCooccurrences);
[*] }
[*] }
Reducer ：SimilarityReducer类，生成协同矩阵
Java代码
[*]public static class SimilarityReducer
[*] extends Reducer<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] @Override
[*] protected void reduce(IntWritable row, Iterable<VectorWritable> partialDots, Context ctx)
[*] throws IOException, InterruptedException {
[*] Iterator<VectorWritable> partialDotsIterator = partialDots.iterator();
[*] //取一个vecotr作为该item的行向量
[*] Vector dots = partialDotsIterator.next().get();
[*] while (partialDotsIterator.hasNext()) {
[*] Vector toAdd = partialDotsIterator.next().get();
[*] Iterator<Vector.Element> nonZeroElements = toAdd.iterateNonZero();
[*] while (nonZeroElements.hasNext()) {
[*] Vector.Element nonZeroElement = nonZeroElements.next();
[*] //nonZeroElement.index()为itemid，将另一个vecotr中itemid的value加进去
[*] dots.setQuick(nonZeroElement.index(), dots.getQuick(nonZeroElement.index()) + nonZeroElement.get());
[*] }
[*] }
[*] //最后得到的dots是协同矩阵中行号为row的一行，行中元素是item对其他的item的相似度
[*] Vector similarities = dots.like();
[*] double normA = norms.getQuick(row.get());
[*] Iterator<Vector.Element> dotsWith = dots.iterateNonZero();
[*] while (dotsWith.hasNext()) {
[*] Vector.Element b = dotsWith.next();
[*] double similarityValue = similarity.similarity(b.get(), normA, norms.getQuick(b.index()), numberOfColumns);
[*] if (similarityValue >= treshold) {
[*] similarities.set(b.index(), similarityValue);
[*] }
[*] }
[*] if (excludeSelfSimilarity) {
[*] similarities.setQuick(row.get(), 0);
[*] }
[*] ctx.write(row, new VectorWritable(similarities));
[*] }
[*] }
3、Mapper：UnsymmetrifyMapper类，用来生成对称矩阵的。上一步得到的是非对称矩阵，首先将矩阵偏转，得到偏转矩阵，用原矩阵加上偏转矩阵，可以得到对称矩阵
Java代码
[*]public static class UnsymmetrifyMapper extends Mapper<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] private int maxSimilaritiesPerRow;
[*]
[*] @Override
[*] protected void setup(Mapper.Context ctx) throws IOException, InterruptedException {
[*] maxSimilaritiesPerRow = ctx.getConfiguration().getInt(MAX_SIMILARITIES_PER_ROW, 0);
[*] Preconditions.checkArgument(maxSimilaritiesPerRow > 0, "Incorrect maximum number of similarities per row!");
[*] }
[*]
[*] @Override
[*] protected void map(IntWritable row, VectorWritable similaritiesWritable, Context ctx)
[*] throws IOException, InterruptedException {
[*] Vector similarities = similaritiesWritable.get();
[*] // For performance reasons moved transposedPartial creation out of the while loop and reusing the same vector
[*] Vector transposedPartial = similarities.like();
[*] TopK<Vector.Element> topKQueue = new TopK<Vector.Element>(maxSimilaritiesPerRow, Vectors.BY_VALUE);
[*] Iterator<Vector.Element> nonZeroElements = similarities.iterateNonZero();
[*] //这个地方用来生成偏转矩阵的，非对称矩阵，用原矩阵加上偏转矩阵，可以得到对称矩阵
[*] while (nonZeroElements.hasNext()) {
[*] Vector.Element nonZeroElement = nonZeroElements.next();
[*] topKQueue.offer(new Vectors.TemporaryElement(nonZeroElement));
[*]
[*] transposedPartial.setQuick(row.get(), nonZeroElement.get());
[*] //偏转矩阵中的每一个元素
[*] ctx.write(new IntWritable(nonZeroElement.index()), new VectorWritable(transposedPartial));
[*]
[*] transposedPartial.setQuick(row.get(), 0.0);
[*] }
[*] Vector topKSimilarities = similarities.like();
[*] for (Vector.Element topKSimilarity : topKQueue.retrieve()) {
[*] topKSimilarities.setQuick(topKSimilarity.index(), topKSimilarity.get());
[*] }
[*] //这里只收集前maxSimilaritiesPerRow个得分最高的item，所以咱们最后的对称矩阵，实际上每行只有
[*] //maxSimilaritiesPerRow个是对称的，其他的位置也不管了
[*] ctx.write(row, new VectorWritable(topKSimilarities));
[*] }
[*] }
Reducer：MergeToTopKSimilaritiesReducer类，就是将上面Map偏转的元素都收集起来，也就是完成了偏转矩阵和（截取了得分前maxSimilaritiesPerRow个）的原矩阵相加的过程，得到了对称矩阵
Java代码
[*]public static class MergeToTopKSimilaritiesReducer
[*] extends Reducer<IntWritable,VectorWritable,IntWritable,VectorWritable> {
[*]
[*] private int maxSimilaritiesPerRow;
[*]
[*] @Override
[*] protected void setup(Context ctx) throws IOException, InterruptedException {
[*] maxSimilaritiesPerRow = ctx.getConfiguration().getInt(MAX_SIMILARITIES_PER_ROW, 0);
[*] Preconditions.checkArgument(maxSimilaritiesPerRow > 0, "Incorrect maximum number of similarities per row!");
[*] }
[*]
[*] @Override
[*] protected void reduce(IntWritable row, Iterable<VectorWritable> partials, Context ctx)
[*] throws IOException, InterruptedException {
[*] Vector allSimilarities = Vectors.merge(partials);
[*] Vector topKSimilarities = Vectors.topKElements(maxSimilaritiesPerRow, allSimilarities);
[*] ctx.write(row, new VectorWritable(topKSimilarities));
[*] }
[*]}
至此，RowSimilarityJob类的全部工作就完成，最终生成的是一个对称矩阵，也就是协同矩阵
Java代码
[*]//协同矩阵与用户向量相乘
[*] //start the multiplication of the co-occurrence matrix by the user vectors
[*] if (shouldRunNextPhase(parsedArgs, currentPhase)) {
[*] //第一个MapReducer
[*] Job prePartialMultiply1 = prepareJob(
[*] similarityMatrixPath, prePartialMultiplyPath1, SequenceFileInputFormat.class,
[*] SimilarityMatrixRowWrapperMapper.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] Reducer.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] SequenceFileOutputFormat.class);
[*] boolean succeeded = prePartialMultiply1.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] //第二个MapReduce
[*] //continue the multiplication
[*] Job prePartialMultiply2 = prepareJob(new Path(prepPath, PreparePreferenceMatrixJob.USER_VECTORS),
[*] prePartialMultiplyPath2, SequenceFileInputFormat.class, UserVectorSplitterMapper.class, VarIntWritable.class,
[*] VectorOrPrefWritable.class, Reducer.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] SequenceFileOutputFormat.class);
[*] if (usersFile != null) {
[*] prePartialMultiply2.getConfiguration().set(UserVectorSplitterMapper.USERS_FILE, usersFile);
[*] }
[*] prePartialMultiply2.getConfiguration().setInt(UserVectorSplitterMapper.MAX_PREFS_PER_USER_CONSIDERED,
[*] maxPrefsPerUser);
[*] succeeded = prePartialMultiply2.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] //finish the job
[*] //第三个MapReduce
[*] Job partialMultiply = prepareJob(
[*] new Path(prePartialMultiplyPath1 + "," + prePartialMultiplyPath2), partialMultiplyPath,
[*] SequenceFileInputFormat.class, Mapper.class, VarIntWritable.class, VectorOrPrefWritable.class,
[*] ToVectorAndPrefReducer.class, VarIntWritable.class, VectorAndPrefsWritable.class,
[*] SequenceFileOutputFormat.class);
[*] setS3SafeCombinedInputPath(partialMultiply, getTempPath(), prePartialMultiplyPath1, prePartialMultiplyPath2);
[*] succeeded = partialMultiply.waitForCompletion(true);
[*] if (!succeeded)
[*] return -1;
[*] }
下边也是同样分析一下这个三个MapReduce的细节：
1、Mapper： SimilarityMatrixRowWrapperMapper 类，将协同矩阵的一行拿出来，通过包装，封装成VectorOrPrefWritable类，与那边的UserVectorSplitterMapper 的输出类型一致
Java代码
[*]public final class SimilarityMatrixRowWrapperMapper extends
[*] Mapper<IntWritable,VectorWritable,VarIntWritable,VectorOrPrefWritable> {
[*]
[*] //将协同矩阵的一行拿出来，通过包装，封装成VectorOrPrefWritable类，与那边的UserVectorSplitterMapper
[*] //的输出类型一致
[*] @Override
[*] protected void map(IntWritable key,
[*] VectorWritable value,
[*] Context context) throws IOException, InterruptedException {
[*] Vector similarityMatrixRow = value.get();
[*] /* remove self similarity */
[*] similarityMatrixRow.set(key.get(), Double.NaN);
[*] context.write(new VarIntWritable(key.get()), new VectorOrPrefWritable(similarityMatrixRow));
[*] }
[*]
[*]}
2、Mapper：UserVectorSplitterMapper类
Java代码
[*]//输入格式： theUserID：<itemid_index1,pref1>,<itemid_index2,pref2>........<itemid_indexN,prefN>
[*] //输出格式： itemid1：<theUserID,pref1>
[*] // itemid2：<theUserID,pref2>
[*] // itemid3：<theUserID,pref3>
[*] // ......
[*] // itemidN：<theUserID,prefN>
Java代码
[*]public final class UserVectorSplitterMapper extends
[*] Mapper<VarLongWritable,VectorWritable, VarIntWritable,VectorOrPrefWritable> {
[*]
[*] @Override
[*] protected void map(VarLongWritable key,
[*] VectorWritable value,
[*] Context context) throws IOException, InterruptedException {
[*] long userID = key.get();
[*] if (usersToRecommendFor != null && !usersToRecommendFor.contains(userID)) {
[*] return;
[*] }
[*] Vector userVector = maybePruneUserVector(value.get());
[*] Iterator<Vector.Element> it = userVector.iterateNonZero();
[*] VarIntWritable itemIndexWritable = new VarIntWritable();
[*] VectorOrPrefWritable vectorOrPref = new VectorOrPrefWritable();
[*] while (it.hasNext()) {
[*] Vector.Element e = it.next();
[*] itemIndexWritable.set(e.index());
[*] vectorOrPref.set(userID, (float) e.get());
[*] context.write(itemIndexWritable, vectorOrPref);
[*] }
[*] }
3、Reduce：ToVectorAndPrefReducer类，收集协同矩阵为itemid的一行，并且收集评价过该item的用户和评分，最后的输出是 itemid_index,VectorAndPrefsWritable(vector,List<userid>,List<pref>)
Java代码
[*]public final class ToVectorAndPrefReducer extends
[*] Reducer<VarIntWritable,VectorOrPrefWritable,VarIntWritable,VectorAndPrefsWritable> {
[*]
[*] //收集所有key为itemid的
[*] @Override
[*] protected void reduce(VarIntWritable key,
[*] Iterable<VectorOrPrefWritable> values,
[*] Context context) throws IOException, InterruptedException {
[*]
[*] List<Long> userIDs = Lists.newArrayList();
[*] List<Float> prefValues = Lists.newArrayList();
[*] Vector similarityMatrixColumn = null;
[*] for (VectorOrPrefWritable value : values) {
[*] if (value.getVector() == null) {
[*] // Then this is a user-pref value
[*] userIDs.add(value.getUserID());
[*] prefValues.add(value.getValue());
[*] } else {
[*] // Then this is the column vector
[*] //协同矩阵的一个行（行号为itemid的一行）
[*] if (similarityMatrixColumn != null) {
[*] throw new IllegalStateException("Found two similarity-matrix columns for item index " + key.get());
[*] }
[*] similarityMatrixColumn = value.getVector();
[*] }
[*] }
[*]
[*] if (similarityMatrixColumn == null) {
[*] return;
[*] }
[*] //收集协同矩阵为itemid的一行，并且手机评价过该item的用户和评分
[*] VectorAndPrefsWritable vectorAndPrefs = new VectorAndPrefsWritable(similarityMatrixColumn, userIDs, prefValues);
[*] context.write(key, vectorAndPrefs);
[*] }
[*]
[*]}
第四步，协同矩阵和用户向量相乘，得到推荐结果
Java代码
[*]//extract out the recommendations
[*] Job aggregateAndRecommend = prepareJob(
[*] new Path(aggregateAndRecommendInput), outputPath, SequenceFileInputFormat.class,
[*] PartialMultiplyMapper.class, VarLongWritable.class, PrefAndSimilarityColumnWritable.class,
[*] AggregateAndRecommendReducer.class, VarLongWritable.class, RecommendedItemsWritable.class,
[*] TextOutputFormat.class);
[*] Configuration aggregateAndRecommendConf = aggregateAndRecommend.getConfiguration();
Mapper：PartialMultiplyMapper类
Java代码
[*]//输入类型：( itemid_index, <userid的数组，pref的数组，协同矩阵行号为itemid_index的行> )
[*]//输出类型： userid，<该用户对itemid_index1的评分，协同矩阵行号为itemid_index1的行> )
[*]// userid，<该用户对itemid_index2的评分，协同矩阵行号为itemid_index2的行> )
[*]// .....
[*]// .....
[*]// userid，<该用户对itemid_indexN的评分，协同矩阵行号为itemid_indexN的行> )
Java代码
[*]public final class PartialMultiplyMapper extends
[*] Mapper<VarIntWritable,VectorAndPrefsWritable,VarLongWritable,PrefAndSimilarityColumnWritable> {
[*]
[*] @Override
[*] protected void map(VarIntWritable key,
[*] VectorAndPrefsWritable vectorAndPrefsWritable,
[*] Context context) throws IOException, InterruptedException {
[*]
[*] Vector similarityMatrixColumn = vectorAndPrefsWritable.getVector();
[*] List<Long> userIDs = vectorAndPrefsWritable.getUserIDs();
[*] List<Float> prefValues = vectorAndPrefsWritable.getValues();
[*]
[*] VarLongWritable userIDWritable = new VarLongWritable();
[*] PrefAndSimilarityColumnWritable prefAndSimilarityColumn = new PrefAndSimilarityColumnWritable();
[*]
[*] for (int i = 0; i < userIDs.size(); i++) {
[*] long userID = userIDs.get(i);
[*] float prefValue = prefValues.get(i);
[*] if (!Float.isNaN(prefValue)) {
[*] prefAndSimilarityColumn.set(prefValue, similarityMatrixColumn);
[*] userIDWritable.set(userID);
[*] context.write(userIDWritable, prefAndSimilarityColumn);
[*] }
[*] }
[*] }
[*]
[*]}
Reducer：AggregateAndRecommendReducer类，Reducer中进行PartialMultiply，按乘积得到的推荐度的大小取出最大的几个item。对于非booleanData，是用pref和相似度矩阵的PartialMultiply得到推荐度的值来进行排序。
而booleanData的pref值都是1.0f，所以去计算矩阵相乘的过程没有意义，直接累加相似度的值即可。
用这个数据排序就可得到推荐结果
Java代码
[*]public final class AggregateAndRecommendReducer extends
[*] Reducer<VarLongWritable,PrefAndSimilarityColumnWritable,VarLongWritable,RecommendedItemsWritable> {
[*] @Override
[*] protected void reduce(VarLongWritable userID,
[*] Iterable<PrefAndSimilarityColumnWritable> values,
[*] Context context) throws IOException, InterruptedException {
[*] if (booleanData) {
[*] reduceBooleanData(userID, values, context);
[*] } else {
[*] reduceNonBooleanData(userID, values, context);
[*] }
[*] }
[*]
[*] private void reduceBooleanData(VarLongWritable userID,
[*] Iterable<PrefAndSimilarityColumnWritable> values,
[*] Context context) throws IOException, InterruptedException {
[*] /* having boolean data, each estimated preference can only be 1,
[*] * however we can't use this to rank the recommended items,
[*] * so we use the sum of similarities for that. */
[*] Vector predictionVector = null;
[*] for (PrefAndSimilarityColumnWritable prefAndSimilarityColumn : values) {
[*] predictionVector = predictionVector == null
[*] ? prefAndSimilarityColumn.getSimilarityColumn()
[*] : predictionVector.plus(prefAndSimilarityColumn.getSimilarityColumn());
[*] }
[*] writeRecommendedItems(userID, predictionVector, context);
[*] }
[*]
[*] private void reduceNonBooleanData(VarLongWritable userID,
[*] Iterable<PrefAndSimilarityColumnWritable> values,
[*] Context context) throws IOException, InterruptedException {
[*] /* each entry here is the sum in the numerator of the prediction formula */
[*] Vector numerators = null;
[*] /* each entry here is the sum in the denominator of the prediction formula */
[*] Vector denominators = null;
[*] /* each entry here is the number of similar items used in the prediction formula */
[*] Vector numberOfSimilarItemsUsed = new RandomAccessSparseVector(Integer.MAX_VALUE, 100);
[*]
[*] for (PrefAndSimilarityColumnWritable prefAndSimilarityColumn : values) {
[*] Vector simColumn = prefAndSimilarityColumn.getSimilarityColumn();
[*] float prefValue = prefAndSimilarityColumn.getPrefValue();
[*] /* count the number of items used for each prediction */
[*] Iterator<Vector.Element> usedItemsIterator = simColumn.iterateNonZero();
[*] while (usedItemsIterator.hasNext()) {
[*] int itemIDIndex = usedItemsIterator.next().index();
[*] numberOfSimilarItemsUsed.setQuick(itemIDIndex, numberOfSimilarItemsUsed.getQuick(itemIDIndex) + 1);
[*] }
[*] //vector.times(float) 是向量乘于一个数，也就是向量的每一个值都乘以这个数
[*] //vector.plus(vector) 是两个向量相加，每一个位置上的值相加
[*]
[*] //numerators是一个vecotr，每一个元素是这样的
[*] /*
[*] 例如index为item1的元素的值为：
[*] simility(item1, item_2)*pref(userid, item_2)
[*] + simility(item_1, item_3)*pref(userid, item_3)
[*] + simility(item1, item_4)*pref(userid, item_4)
[*] + ……
[*] + simility(item_1, item_2)*pref(userid, item_N)
[*] */
[*] // 注：其中simility(item1, item2)代表物品item1和物品item2的相似度 ，pref(userid, item)代表用于userid对item打分分值
[*]
[*] numerators = numerators == null
[*] ? prefValue == BOOLEAN_PREF_VALUE ? simColumn.clone() : simColumn.times(prefValue)
[*] : numerators.plus(prefValue == BOOLEAN_PREF_VALUE ? simColumn : simColumn.times(prefValue));
[*]
[*]
[*]
[*] simColumn.assign(ABSOLUTE_VALUES);
[*] //denominators是一个vecotr，每一个元素是这样的
[*] /*
[*] 例如index为item1的元素的值为：
[*] simility(item1, item_2)+ simility(item_1, item_3)+ …… + simility(item_1, item_2)*pref(userid, item_N)
[*] */
[*] // 注：其中simility(item1, item2)代表物品item1和物品item2的相似度
[*] denominators = denominators == null ? simColumn : denominators.plus(simColumn);
[*] }
[*]
[*] if (numerators == null) {
[*] return;
[*] }
[*]
[*] Vector recommendationVector = new RandomAccessSparseVector(Integer.MAX_VALUE, 100);
[*] Iterator<Vector.Element> iterator = numerators.iterateNonZero();
[*] while (iterator.hasNext()) {
[*] Vector.Element element = iterator.next();
[*] int itemIDIndex = element.index();
[*] /* preference estimations must be based on at least 2 datapoints */
[*] if (numberOfSimilarItemsUsed.getQuick(itemIDIndex) > 1) {
[*] /* compute normalized prediction */
[*] //计算归一化预测值
[*] double prediction = element.get() / denominators.getQuick(itemIDIndex);
[*] recommendationVector.setQuick(itemIDIndex, prediction);
[*] }
[*] }
[*] writeRecommendedItems(userID, recommendationVector, context);
[*] }
[*]}
页:
[1]