在使用Pipeline串联多个stage时model和非model的区别

train.csv数据：

id,name,age,sex
1,lyy,20,F
2,rdd,20,M
3,nyc,18,M
4,mzy,10,M

数据读取：

 SparkSession  spark = SparkSession.builder().enableHiveSupport()
                     .getOrCreate();
         Dataset<Row> dataset = spark
                 .read()
                 .format("org.apache.spark.sql.execution.datasources.csv.CSVFileFormat")
                 .option("header", true)
                 .option("inferSchema", true)
                 .option("delimiter", ",")
                 //.load("file:///E:/git/bigdata_sparkIDE/spark-ide/workspace/test/SparkMLTest/SanFranciscoCrime/document/kaggle-旧金山犯罪分类/train-new.csv") //PreProcess1
                 .load("file:///E:/git/bigdata_sparkIDE/spark-ide/workspace/test/SparkMLTest/DataPreprocessing/document/train.csv") //PreProcess2
                 .persist();

     public static void PreProcess2(Dataset<Row> data) {

                 data.printSchema();
                 // 重新索引标签值
                 StringIndexerModel labelIndexer = new StringIndexer()
                 .setInputCol("sex")
                 .setOutputCol("label")
                 .fit(data);

                 StringIndexerModel nameIndexer = new StringIndexer()
                 .setInputCol("name")
                 .setOutputCol("namenum")
                 .fit(data);

                 /*  会报错：Exception in thread "main" java.lang.IllegalArgumentException: Field "namenum" does not exist.
                  * 原因是：Model类型调用fit时，要求数据集中必须包含InputCol所指定的列名
                  * 不会将Pipeline某个stage的输出作为InputCol，即使那个stage的OutputCol指定的列名与其相同也不行
                  * StringIndexerModel name1Indexer = new StringIndexer()
                 .setInputCol("namenum")
                 .setOutputCol("namenum1")
                 .fit(data);*/

                 /* 错误原因StringIndexerModel错误一样，features并不是data的列
                  * VectorIndexerModel featureIndexer = new VectorIndexer()
                     .setInputCol("features")
                     .setOutputCol("indexfeatures")
                     .setMaxCategories(4)
                     .fit(data);*/

                 //成功
                 //原因说明：非model时，转换器不会调用fit,而会使用Pipeline某个stage的输出作为InputCol
                 //由于stage[2]即 assembler已经生成features，故而该处直接使用；
                 //但是该类型时不能单独使用，必须依赖Pipeline
                 VectorIndexer featureIndexer = new VectorIndexer()
                 .setInputCol("features")
                 .setOutputCol("indexfeatures")
                 .setMaxCategories();

                 //由上述分析可知，该处输入的列可以是多个stage的输出组成，因为VectorAssembler非model
                 //因此可以使用中间生成结果，且可以使用多个
                 VectorAssembler assembler = new VectorAssembler()
                 .setInputCols("id,namenum,age".split(","))
                .setOutputCol("features");

                 //这里的stage的顺序很重要，一定按照依赖关系顺序放入，如下顺序就会报错：
                 //Exception in thread "main" java.lang.IllegalArgumentException: Field "features" does not exist.
                 //Pipeline pipeline = new Pipeline().setStages(new PipelineStage[] {labelIndexer,nameIndexer,featureIndexer,assembler});

                 //将featureIndexer放到assembler即可
                 Pipeline pipeline = new Pipeline().setStages(new PipelineStage[] {labelIndexer,nameIndexer,assembler,featureIndexer});

                 // Train model. This also runs the indexers.
                 PipelineModel model = pipeline.fit(data);

                 // Make predictions.
                 Dataset<Row> result = model.transform(data);

                 result.show(, false);

     }

root
|-- id: integer (nullable = true)
|-- name: string (nullable = true)
|-- age: integer (nullable = true)
|-- sex: string (nullable = true)

+---+----+---+---+-----+-------+--------------+-------------+
|id |name|age|sex|label|namenum|features |indexfeatures|
+---+----+---+---+-----+-------+--------------+-------------+
|1 |lyy |20 |F |1.0 |1.0 |[1.0,1.0,20.0]|[0.0,1.0,2.0]|
|2 |rdd |20 |M |0.0 |2.0 |[2.0,2.0,20.0]|[1.0,2.0,2.0]|
|3 |nyc |18 |M |0.0 |0.0 |[3.0,0.0,18.0]|[2.0,0.0,1.0]|
|4 |mzy |10 |M |0.0 |3.0 |[4.0,3.0,10.0]|[3.0,3.0,0.0]|
+---+----+---+---+-----+-------+--------------+-------------+

综上分析，可以将原有代码做一简化：

 public static void PreProcess2(Dataset<Row> data) {

                 data.printSchema();
                 // 重新索引标签值
                 StringIndexer labelIndexer = new StringIndexer()
                 .setInputCol("sex")
                 .setOutputCol("label");

                 StringIndexer nameIndexer = new StringIndexer()
                 .setInputCol("name")
                 .setOutputCol("namenum");

                 VectorIndexer featureIndexer = new VectorIndexer()
                 .setInputCol("features")
                 .setOutputCol("indexfeatures")
                 .setMaxCategories();

                 VectorAssembler assembler = new VectorAssembler()
                 .setInputCols("id,namenum,age".split(","))
                .setOutputCol("features");

                 Pipeline pipeline = new Pipeline().setStages(new PipelineStage[] {labelIndexer,nameIndexer,assembler,featureIndexer});

                 // Train model. This also runs the indexers.
                 PipelineModel model = pipeline.fit(data);  //以这里的data为基准数据

                 // Make predictions.
                 Dataset<Row> result = model.transform(data);

                 result.show(, false);

     }

运行结果：

root
 |-- id: integer (nullable = true)
 |-- name: string (nullable = true)
 |-- age: integer (nullable = true)
 |-- sex: string (nullable = true)

+---+----+---+---+-----+-------+--------------+-------------+
|id |name|age|sex|label|namenum|features      |indexfeatures|
+---+----+---+---+-----+-------+--------------+-------------+
|  |lyy | |F  |1.0  |1.0    |[1.0,1.0,20.0]|[0.0,1.0,2.0]|
|  |rdd | |M  |0.0  |2.0    |[2.0,2.0,20.0]|[1.0,2.0,2.0]|
|  |nyc | |M  |0.0  |0.0    |[3.0,0.0,18.0]|[2.0,0.0,1.0]|
|  |mzy | |M  |0.0  |3.0    |[4.0,3.0,10.0]|[3.0,3.0,0.0]|
+---+----+---+---+-----+-------+--------------+-------------+