java - 使用Java的Spark MLlib分类输入格式

Question

如何将 DTO 列表转换为 Spark ML 输入数据集格式？

我有 DTO:

public class MachineLearningDTO implements Serializable {
    private double label;
    private double[] features;

    public MachineLearningDTO() {
    }

    public MachineLearningDTO(double label, double[] features) {
        this.label = label;
        this.features = features;
    }

    public double getLabel() {
        return label;
    }

    public void setLabel(double label) {
        this.label = label;
    }

    public double[] getFeatures() {
        return features;
    }

    public void setFeatures(double[] features) {
        this.features = features;
    }
}

和代码:

Dataset<MachineLearningDTO> mlInputDataSet = spark.createDataset(mlInputData, Encoders.bean(MachineLearningDTO.class));
LogisticRegression logisticRegression = new LogisticRegression();
LogisticRegressionModel model = logisticRegression.fit(MLUtils.convertMatrixColumnsToML(mlInputDataSet));

执行代码后我得到:

java.lang.IllegalArgumentException: requirement failed: Column features must be of type org.apache.spark.ml.linalg.VectorUDT@3bfc3ba7 but was actually ArrayType(DoubleType,false).

如果将其更改为org.apache.spark.ml.linalg.VectorUDT，代码如下:

VectorUDT vectorUDT = new VectorUDT();
vectorUDT.serialize(Vectors.dense(......));

然后我得到:

java.lang.UnsupportedOperationException: Cannot infer type for class org.apache.spark.ml.linalg.VectorUDT because it is not bean-compliant

at org.apache.spark.sql.catalyst.JavaTypeInference$.org$apache$spark$sql$catalyst$JavaTypeInference$$serializerFor(JavaTypeInference.scala:437)

Answer 1

我已经想通了，以防万一有人也坚持使用它，我编写了简单的转换器并且它可以工作:

private Dataset<Row> convertToMlInputFormat(List< MachineLearningDTO> data) {
    List<Row> rowData = data.stream()
            .map(dto ->
                    RowFactory.create(dto.getLabel() ? 1.0d : 0.0d, Vectors.dense(dto.getFeatures())))
            .collect(Collectors.toList());
    StructType schema = new StructType(new StructField[]{
            new StructField("label", DataTypes.DoubleType, false, Metadata.empty()),
            new StructField("features", new VectorUDT(), false, Metadata.empty()),
    });

    return spark.createDataFrame(rowData, schema);
}