Weka進階應用程式--Java API__Java

1.引入

本文是我學習《資料採礦與機器學習–WEKA應用技術與實踐》的筆記。該書電子版的連結是：http://download.csdn.net/detail/fhb292262794/8759397
前一篇博文總結了用Weka示範機器學習的演算法處理，主要是通過Weka3.8的用戶端軟體操作。
本文通過Java API調用來處理，這樣就可以在編程中應用Weka的機器學習演算法處理資料。

本書的執行個體是用weka3.7，我下載使用的是最新版的weka3.8，更新代碼以適應了weka3.8後，整理記錄如下。 1.分類（手把手教你寫代碼） 1.1 線性迴歸

預測樓價

樓價資料：

@RELATION house@ATTRIBUTE houseSize NUMERIC@ATTRIBUTE lotSize NUMERIC@ATTRIBUTE bedrooms NUMERIC@ATTRIBUTE granite NUMERIC@ATTRIBUTE bathroom NUMERIC@ATTRIBUTE sellingPrice NUMERIC@DATA3529,9191,6,0,0,205000 3247,10061,5,1,1,224900 4032,10150,5,0,1,197900 2397,14156,4,1,0,189900 2200,9600,4,0,1,195000 3536,19994,6,1,1,325000 2983,9365,5,0,1,230000 

需求：根據該地區附近的房子資訊及樓價出售價格，預測新的房子的售價。該房子資訊是：houseSize:3198;lotSize:9669;bedrooms:5;granite:3;bathroom:1;請預測樓價。

由需求可知，需要處理的邏輯是：
1.載入樓價資料。
2.設定屬性資訊。
3.構建分類器並計算係數。
4.使用迴歸係數預測未知樓價。

代碼如下：

    public static final String WEKA_PATH = "data/weka/";    public static final String WEATHER_NOMINAL_PATH = "data/weka/weather.nominal.arff";    public static final String WEATHER_NUMERIC_PATH = "data/weka/weather.numeric.arff";    public static final String SEGMENT_CHALLENGE_PATH = "data/weka/segment-challenge.arff";    public static final String SEGMENT_TEST_PATH = "data/weka/segment-test.arff";    public static final String IONOSPHERE_PATH = "data/weka/ionosphere.arff";    public static void pln(String str) {        System.out.println(str);    }    @Test    public void testLinearRegression() throws Exception {        Instances dataset = ConverterUtils.DataSource.read(WEKA_PATH + "houses.arff");        dataset.setClassIndex(dataset.numAttributes() - 1);        LinearRegression linearRegression = new LinearRegression();        try {            linearRegression.buildClassifier(dataset);        } catch (Exception e) {            e.printStackTrace();        }        double[] coef = linearRegression.coefficients();        double myHouseValue = (coef[0] * 3198) +                (coef[1] * 9669) +                (coef[2] * 5) +                (coef[3] * 3) +                (coef[4] * 1) +                coef[6];        System.out.println(myHouseValue);    }

1.2 隨機森林

代碼：

@Test    public void testRandomForestClassifier() throws Exception {        ArffLoader loader = new ArffLoader();        loader.setFile(new File(WEKA_PATH + "segment-challenge.arff"));        Instances instances = loader.getDataSet();        instances.setClassIndex(instances.numAttributes() - 1);        System.out.println(instances);        System.out.println("------------");        RandomForest rf = new RandomForest();        rf.buildClassifier(instances);        System.out.println(rf);    }

1.3 元分類器

// 元分類器    @Test    public void testMetaClassifier() throws Exception {        Instances data = ConverterUtils.DataSource.read(WEATHER_NUMERIC_PATH);        if (data.classIndex() == -1)            data.setClassIndex(data.numAttributes() - 1);        AttributeSelectedClassifier classifier = new AttributeSelectedClassifier();        CfsSubsetEval eval = new CfsSubsetEval();        GreedyStepwise stepwise = new GreedyStepwise();        stepwise.setSearchBackwards(true);        J48 base = new J48();        classifier.setClassifier(base);        classifier.setEvaluator(eval);        classifier.setSearch(stepwise);        Evaluation evaluation = new Evaluation(data);        evaluation.crossValidateModel(classifier, data, 10, new Random(1234));        pln(evaluation.toSummaryString());    }

1.4 預測分類結果（批量處理）

代碼:

 /**     * 利用訓練集預測測試集的分類，批量處理     */    @Test    public void testOutputClassDistribution() throws Exception {        ArffLoader loader = new ArffLoader();        loader.setFile(new File(SEGMENT_CHALLENGE_PATH));        Instances train = loader.getDataSet();        train.setClassIndex(train.numAttributes() - 1);        ArffLoader loader1 = new ArffLoader();        loader1.setFile(new File(SEGMENT_TEST_PATH));        Instances test = loader1.getDataSet();        test.setClassIndex(test.numAttributes() - 1);        J48 classifier = new J48();        classifier.buildClassifier(train);        System.out.println("num\t-\tfact\t-\tpred\t-\terr\t-\tdistribution");        for (int i = 0; i < test.numInstances(); i++) {            double pred = classifier.classifyInstance(test.instance(i));            double[] dist = classifier.distributionForInstance(test.instance(i));            StringBuilder sb = new StringBuilder();            sb.append(i + 1)                    .append(" - ")                    .append(test.instance(i).toString(test.classIndex()))                    .append(" - ")                    .append(test.classAttribute().value((int) pred))                    .append(" - ");            if (pred != test.instance(i).classValue())                sb.append("yes");            else                sb.append("no");            sb.append(" - ");            sb.append(Utils.arrayToString(dist));            System.out.println(sb.toString());        }    }

這裡指定的是J48，是決策樹分類器，可以用其他更好的分類器替代，請比較效果選用分類器。 1.5 交叉驗證

代碼:

// 交叉驗證並預測    @Test    public void testOnceCVAndPrediction() throws Exception {        Instances data = ConverterUtils.DataSource.read(IONOSPHERE_PATH);        data.setClassIndex(data.numAttributes() - 1);        Classifier classifier = new J48();        int seed = 1234;        int folds = 10;        Debug.Random random = new Debug.Random(seed);        Instances newData = new Instances(data);        newData.randomize(random);        if (newData.classAttribute().isNominal())            newData.stratify(folds);        // 執行交叉驗證，並添加預測        Instances predictedData = null;        Evaluation eval = new Evaluation(newData);        for (int i = 0; i < folds; i++) {            Instances train = newData.trainCV(folds, i);            Instances test = newData.testCV(folds, i);            Classifier clsCopy = AbstractClassifier.makeCopy(classifier);            clsCopy.buildClassifier(train);            eval.evaluateModel(clsCopy, test);            // add prediction            AddClassification filter = new AddClassification();            filter.setClassifier(classifier);            filter.setOutputClassification(true);            filter.setOutputDistribution(true);            filter.setOutputErrorFlag(true);            filter.setInputFormat(train);            Filter.useFilter(train, filter);            Instances pred = Filter.useFilter(test, filter);            if (predictedData == null)                predictedData = new Instances(pred, 0);            for (int j = 0; j < pred.numInstances(); j++)                predictedData.add(pred.instance(j));        }        pln("classifier:" + classifier.getClass().getName() + " " + Utils.joinOptions(((OptionHandler) classifier).getOptions()));        pln("data:" + data.relationName());        pln("seed:" + seed);        pln(eval.toSummaryString("=== " + folds + " test ===", false));        // write data        ConverterUtils.DataSink.write(WEKA_PATH + "predictions.arff", predictedData);    }

2.聚類（手把手教你寫代碼） 2.1 EM

@Test    public void testEM() throws Exception {        Instances instances = ConverterUtils.DataSource.read(WEKA_PATH + "contact-lenses.arff");        EM cluster = new EM();        cluster.setOptions(new String[]{"-I", "100"});        cluster.buildClusterer(instances);        pln(cluster.toString());    }

2.2 估計聚類器

// 評估聚類器的方式 3種    @Test    public void testEvaluation() throws Exception {        String filePath = WEKA_PATH + "contact-lenses.arff";        Instances instances = ConverterUtils.DataSource.read(filePath);        // 第1種        String[] options = new String[]{"-t", filePath};        String output = ClusterEvaluation.evaluateClusterer(new EM(), options);        pln(output);        // 第2種        DensityBasedClusterer dbc = new EM();        dbc.buildClusterer(instances);        ClusterEvaluation clusterEvaluation = new ClusterEvaluation();        clusterEvaluation.setClusterer(dbc);        clusterEvaluation.evaluateClusterer(new Instances(instances));        pln(clusterEvaluation.clusterResultsToString());        // 第3種        // 基於密度的聚類器交叉驗證        DensityBasedClusterer newdbc = new EM();        double logLikelyhood = ClusterEvaluation.crossValidateModel(newdbc, instances, 10, instances.getRandomNumberGenerator(1234));        pln("logLikelyhood: " + logLikelyhood);    }

2.3 聚類並評估

@Test    public void testClassesToClusters() throws Exception {        String filePath = WEKA_PATH + "contact-lenses.arff";        Instances data = ConverterUtils.DataSource.read(filePath);        data.setClassIndex(data.numAttributes() - 1);        Remove remove = new Remove();        remove.setAttributeIndices("" + (data.classIndex() + 1));        remove.setInputFormat(data);        Instances dataCluster = Filter.useFilter(data, remove);        Clusterer cluster = new EM();        cluster.buildClusterer(dataCluster);        ClusterEvaluation eval = new ClusterEvaluation();        eval.setClusterer(cluster);        eval.evaluateClusterer(data);        pln(eval.clusterResultsToString());    }

2.4 輸出聚類點

@Test    public void testOutputClusterDistribution() throws Exception {        Instances train = ConverterUtils.DataSource.read(SEGMENT_CHALLENGE_PATH);        Instances test = ConverterUtils.DataSource.read(SEGMENT_TEST_PATH);        if (!train.equalHeaders(test))            throw new Exception("train data and test data not the same headers.");        EM clusterer = new EM();        clusterer.buildClusterer(train);        pln("id - cluster - distribution");        for (int i = 0; i < test.numInstances(); i++) {            int cluster = clusterer.clusterInstance(test.instance(i));            double[] dists = clusterer.distributionForInstance(test.instance(i));            StringBuilder sb = new StringBuilder();            sb.append(i + 1).append(" - ").append(cluster).append(" - ").append(Utils.arrayToString(dists));            pln(sb.toString());        }    }

3.屬性選擇（手把手教你寫代碼） 自動屬性選擇

應用CfsSubsetEval及GreedyStepwise處理：

// 底層API屬性選擇    @Test    public void testUseLowApi() throws Exception {        ConverterUtils.DataSource source = new ConverterUtils.DataSource(WEATHER_NOMINAL_PATH);        Instances data = source.getDataSet();        if(data.classIndex() == -1)            data.setClassIndex(data.numAttributes() -1);        AttributeSelection attributeSelection = new AttributeSelection();        CfsSubsetEval eval = new CfsSubsetEval();        GreedyStepwise search = new GreedyStepwise();        search.setSearchBackwards(true);        attributeSelection.setEvaluator(eval);        attributeSelection.setSearch(search);        attributeSelection.SelectAttributes(data);        int[] indices = attributeSelection.selectedAttributes();        pln(Utils.arrayToString(indices));    }

4.其他 4.1 資料庫表操作

@Test    public void testSaveCSV() throws Exception {        DatabaseLoader loader = new DatabaseLoader();        loader.setUrl(SqlUtil.URL);        loader.setUser(SqlUtil.USER);        loader.setPassword(SqlUtil.PASSWORD);        loader.setQuery("select question from question");        Instances data1 = loader.getDataSet();        if (data1.classIndex() == -1)            data1.setClassIndex(data1.numAttributes() - 1);        System.out.println(data1);        CSVSaver saver = new CSVSaver();        saver.setInstances(data1);        saver.setFile(new File("data/weka/baidubook-csvsaver.csv"));        saver.writeBatch();    }

4.2 過濾器

過濾

@Test    public void testFilter() throws Exception {        Instances instances = ConverterUtils.DataSource.read("data/weka/houses.arff");        instances.setClassIndex(instances.numAttributes() - 1);        System.out.println(instances);        String[] options = new String[2];        options[0]  = "-R";        options[1] = "1";        Remove remove = new Remove();        remove.setOptions(options);        remove.setInputFormat(instances);        Instances newData = Filter.useFilter(instances,remove);        System.out.println(newData);    }

過濾並分類

@Test    public void testFilterOnTheFly() throws Exception {        Instances instances = ConverterUtils.DataSource.read("data/weka/weather.nominal.arff");        instances.setClassIndex(instances.numAttributes() - 1);        System.out.println(instances);        Remove remove = new Remove();        remove