Examples with EvalBinaryClassBatchOp com.alibaba.alink.operator.batch.evaluation.EvalBinaryClassBatchOp used on opensource projects

Example 21 with EvalBinaryClassBatchOp

use of com.alibaba.alink.operator.batch.evaluation.EvalBinaryClassBatchOp in project Alink by alibaba.

the class Chap11 method c_4.

static void c_4() throws Exception {
    AkSourceBatchOp all_data = new AkSourceBatchOp().setFilePath(DATA_DIR + FEATURE_LABEL_FILE);
    all_data.lazyPrintStatistics().groupBy("label", "label, COUNT(*) AS cnt").print();
    Utils.splitTrainTestIfNotExist(all_data, DATA_DIR + TRAIN_FILE, DATA_DIR + TEST_FILE, 0.8);
    AkSourceBatchOp train_data = new AkSourceBatchOp().setFilePath(DATA_DIR + TRAIN_FILE);
    AkSourceBatchOp test_data = new AkSourceBatchOp().setFilePath(DATA_DIR + TEST_FILE);
    String[] featureColNames = ArrayUtils.removeElement(train_data.getColNames(), LABEL_COL_NAME);
    new LogisticRegression().setFeatureCols(featureColNames).setLabelCol(LABEL_COL_NAME).setPredictionCol(PREDICTION_COL_NAME).setPredictionDetailCol(PRED_DETAIL_COL_NAME).fit(train_data).transform(test_data).link(new EvalBinaryClassBatchOp().setLabelCol(LABEL_COL_NAME).setPredictionDetailCol(PRED_DETAIL_COL_NAME).lazyPrintMetrics("LogisticRegression"));
    BatchOperator.execute();
    if (!new File(DATA_DIR + TRAIN_SAMPLE_FILE).exists()) {
        train_data.link(new StratifiedSampleBatchOp().setStrataRatios("0:0.05,1:1.0").setStrataCol(LABEL_COL_NAME)).link(new AkSinkBatchOp().setFilePath(DATA_DIR + TRAIN_SAMPLE_FILE));
        BatchOperator.execute();
    }
    AkSourceBatchOp train_sample = new AkSourceBatchOp().setFilePath(DATA_DIR + TRAIN_SAMPLE_FILE);
    new LogisticRegression().setFeatureCols(featureColNames).setLabelCol(LABEL_COL_NAME).setPredictionCol(PREDICTION_COL_NAME).setPredictionDetailCol(PRED_DETAIL_COL_NAME).fit(train_sample).transform(test_data).link(new EvalBinaryClassBatchOp().setLabelCol(LABEL_COL_NAME).setPredictionDetailCol(PRED_DETAIL_COL_NAME).lazyPrintMetrics("LogisticRegression with Stratified Sample"));
    BatchOperator.execute();
}

Example 22 with EvalBinaryClassBatchOp

use of com.alibaba.alink.operator.batch.evaluation.EvalBinaryClassBatchOp in project Alink by alibaba.

the class Chap20 method c_1.

static void c_1() throws Exception {
    BatchOperator<?> train_data = new AkSourceBatchOp().setFilePath(Chap10.DATA_DIR + Chap10.TRAIN_FILE).select(Chap10.CLAUSE_CREATE_FEATURES);
    BatchOperator<?> test_data = new AkSourceBatchOp().setFilePath(Chap10.DATA_DIR + Chap10.TEST_FILE).select(Chap10.CLAUSE_CREATE_FEATURES);
    final String[] new_features = ArrayUtils.removeElement(train_data.getColNames(), Chap10.LABEL_COL_NAME);
    LogisticRegression lr = new LogisticRegression().setFeatureCols(new_features).setLabelCol(Chap10.LABEL_COL_NAME).setPredictionCol(Chap10.PREDICTION_COL_NAME).setPredictionDetailCol(Chap10.PRED_DETAIL_COL_NAME);
    Pipeline pipeline = new Pipeline().add(lr);
    GridSearchCV gridSearch = new GridSearchCV().setNumFolds(5).setEstimator(pipeline).setParamGrid(new ParamGrid().addGrid(lr, LogisticRegression.L_1, new Double[] { 0.0000001, 0.000001, 0.00001, 0.0001, 0.001, 0.01, 0.1, 1.0, 10.0 })).setTuningEvaluator(new BinaryClassificationTuningEvaluator().setLabelCol(Chap10.LABEL_COL_NAME).setPredictionDetailCol(Chap10.PRED_DETAIL_COL_NAME).setTuningBinaryClassMetric(TuningBinaryClassMetric.AUC)).enableLazyPrintTrainInfo();
    GridSearchCVModel bestModel = gridSearch.fit(train_data);
    bestModel.transform(test_data).link(new EvalBinaryClassBatchOp().setPositiveLabelValueString("2").setLabelCol(Chap10.LABEL_COL_NAME).setPredictionDetailCol(Chap10.PRED_DETAIL_COL_NAME).lazyPrintMetrics("GridSearchCV"));
    BatchOperator.execute();
}

Example 23 with EvalBinaryClassBatchOp

use of com.alibaba.alink.operator.batch.evaluation.EvalBinaryClassBatchOp in project Alink by alibaba.

the class Chap20 method c_2.

static void c_2() throws Exception {
    Stopwatch sw = new Stopwatch();
    sw.start();
    AlinkGlobalConfiguration.setPrintProcessInfo(true);
    BatchOperator train_sample = new AkSourceBatchOp().setFilePath(Chap11.DATA_DIR + Chap11.TRAIN_SAMPLE_FILE);
    BatchOperator test_data = new AkSourceBatchOp().setFilePath(Chap11.DATA_DIR + Chap11.TEST_FILE);
    final String[] featuresColNames = ArrayUtils.removeElement(train_sample.getColNames(), Chap11.LABEL_COL_NAME);
    GbdtClassifier gbdt = new GbdtClassifier().setFeatureCols(featuresColNames).setLabelCol(Chap11.LABEL_COL_NAME).setPredictionCol(Chap11.PREDICTION_COL_NAME).setPredictionDetailCol(Chap11.PRED_DETAIL_COL_NAME);
    RandomSearchTVSplit randomSearch = new RandomSearchTVSplit().setNumIter(20).setTrainRatio(0.8).setEstimator(gbdt).setParamDist(new ParamDist().addDist(gbdt, GbdtClassifier.NUM_TREES, ValueDist.randArray(new Integer[] { 50, 100 })).addDist(gbdt, GbdtClassifier.MAX_DEPTH, ValueDist.randInteger(4, 10)).addDist(gbdt, GbdtClassifier.MAX_BINS, ValueDist.randArray(new Integer[] { 64, 128, 256, 512 })).addDist(gbdt, GbdtClassifier.LEARNING_RATE, ValueDist.randArray(new Double[] { 0.3, 0.1, 0.01 }))).setTuningEvaluator(new BinaryClassificationTuningEvaluator().setLabelCol(Chap11.LABEL_COL_NAME).setPredictionDetailCol(Chap11.PRED_DETAIL_COL_NAME).setTuningBinaryClassMetric(TuningBinaryClassMetric.F1)).enableLazyPrintTrainInfo();
    RandomSearchTVSplitModel bestModel = randomSearch.fit(train_sample);
    bestModel.transform(test_data).link(new EvalBinaryClassBatchOp().setPositiveLabelValueString("1").setLabelCol(Chap11.LABEL_COL_NAME).setPredictionDetailCol(Chap11.PRED_DETAIL_COL_NAME).lazyPrintMetrics());
    BatchOperator.execute();
    sw.stop();
    System.out.println(sw.getElapsedTimeSpan());
}