sparkML原始数据转换成label-features方法
数据1：kaggle-旧金山犯罪分类数据 格式如下： Dates,Category,Descript,DayOfWeek,PdDistrict,Resolution,Address,X,Y 2015-05-13 23:53:00,WARRANTS,WARRANT ARREST,Wednesday,NORTHERN,"ARREST, BOOKED",OAK ST / LAGUNA ST,-122.425891675136,37.7745985956747 2015-05-13 23:53:00,OTHER OFFENSES,TRAFFIC VIOLATION ARREST,Wednesday,NORTHERN,"ARREST, BOOKED",OAK ST / LAGUNA ST,-122.425891675136,37.7745985956747 2015-05-13 23:33:00,OTHER OFFENSES,TRAFFIC VIOLATION ARREST,Wednesday,NORTHERN,"ARREST, BOOKED",VANNESS AV / GREENWICH ST,-122.42436302145,37.8004143219856 2015-05-13 23:30:00,LARCENY/THEFT,GRAND THEFT FROM LOCKED AUTO,Wednesday,NORTHERN,NONE,1500 Block of LOMBARD ST,-122.42699532676599,37.80087263276921 2015-05-13 23:30:00,LARCENY/THEFT,GRAND THEFT FROM LOCKED AUTO,Wednesday,PARK,NONE,100 Block of BRODERICK ST,-122.438737622757,37.771541172057795 2015-05-13 23:30:00,LARCENY/THEFT,GRAND THEFT FROM UNLOCKED AUTO,Wednesday,INGLESIDE,NONE,0 Block of TEDDY AV,-122.40325236121201,37.713430704116 2015-05-13 23:30:00,VEHICLE THEFT,STOLEN AUTOMOBILE,Wednesday,INGLESIDE,NONE,AVALON AV / PERU AV,-122.423326976668,37.7251380403778 2015-05-13 23:30:00,VEHICLE THEFT,STOLEN AUTOMOBILE,Wednesday,BAYVIEW,NONE,KIRKWOOD AV / DONAHUE ST,-122.371274317441,37.7275640719518 2015-05-13 23:00:00,LARCENY/THEFT,GRAND THEFT FROM LOCKED AUTO,Wednesday,RICHMOND,NONE,600 Block of 47TH AV,-122.508194031117,37.776601260681204 测试代码： public static void main(String[] args) { 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") .persist(); DataPreProcess(dataset); } //此函数包含StringIndexer，OneHotEncoder，VectorAssembler,VectorIndexer数据转换方法 public static Dataset<Row> DataPreProcess(Dataset<Row> data) { //Dataset<Row> df = data.selectExpr("cast(Dates as String) ,DayOfWeek,PdDistrict,Category".split(",")); Dataset<Row> df = data.select(data.col("Dates").cast("String").alias("Dates"),data.col("DayOfWeek").alias("DayOfWeek"),data.col("PdDistrict"),data.col("Category")); df.printSchema(); // 重新索引标签值 SparkLog.info(data.select("Category").distinct().count()); //将非数字类型标签转换成数字类型，按照标签去重的个数n,编号0~n,相同标签的多行记录转换后的数字标签编号相同 //这个适合所有非数字且不连续的有限类别数据编号，不仅仅是只能编号标签 StringIndexerModel labelIndexer = new StringIndexer() .setInputCol("Category").setOutputCol("label").fit(df); StringIndexerModel DateIndexer = new StringIndexer() .setInputCol("Dates").setOutputCol("DatesNum").fit(df); StringIndexerModel DayOfWeekIndexer = new StringIndexer() .setInputCol("DayOfWeek").setOutputCol("dfNum").fit(df); StringIndexerModel PdDistrictIndexer = new StringIndexer() .setInputCol("PdDistrict").setOutputCol("pdNum").fit(df); /*独热编码将类别特征（离散的，已经转换为数字编号形式（这个是必须的，否则会报错）， 映射成独热编码，生成的是一个稀疏向量 比如字符串"abcab"的映射规则:去重后的特征个数n即为稀疏向量的维数，而数字编号代 表该特征对应的向量中非0值的下标，最后生成0-1编码的向量 a 1 0 0 b 0 1 0 c 0 0 1 a 1 0 0 b 0 1 0 */ //OneHotEncoder不需要fit OneHotEncoder encoder = new OneHotEncoder().setInputCol("dfNum") .setOutputCol("dfvec") .setDropLast(false); // 设置最后一个是否包含 OneHotEncoder encoder1 = new OneHotEncoder().setInputCol("pdNum") .setOutputCol("pdvec") .setDropLast(false);// 设置最后一个是否包含 OneHotEncoder encoder2 = new OneHotEncoder().setInputCol("DatesNum") .setOutputCol("Datesvec") .setDropLast(false);// 设置最后一个是否包含 //将多个列拼接成一个向量，列的类型可以是向量 VectorAssembler assembler = new VectorAssembler().setInputCols( "Datesvec,dfvec,pdvec".split(",")).setOutputCol("features"); // Dataset<Row> assembledFeatures = assembler.transform(df); Pipeline pipeline = new Pipeline().setStages(new PipelineStage[] { DateIndexer, DayOfWeekIndexer, PdDistrictIndexer, encoder, encoder1, encoder2, labelIndexer, assembler }); // Train model. This also runs the indexers. PipelineModel model = pipeline.fit(df); // Make predictions. Dataset<Row> predictions = model.transform(df); predictions.describe("label").show(); predictions.show(100, false); return predictions; } +-------------------+---------+----------+--------------+--------+-----+-----+-------------+--------------+-----------------------+-----+---------------------------------------------+ |Dates |DayOfWeek|PdDistrict|Category |DatesNum|dfNum|pdNum|dfvec |pdvec |Datesvec |label|features | +-------------------+---------+----------+--------------+--------+-----+-----+-------------+--------------+-----------------------+-----+---------------------------------------------+ |2015-05-13 23:53:00|Wednesday|NORTHERN |WARRANTS |172231.0|1.0 |2.0 |(7,[1],[1.0])|(10,[2],[1.0])|(389257,[172231],[1.0])|7.0 |(389274,[172231,389258,389266],[1.0,1.0,1.0])| |2015-05-13 23:53:00|Wednesday|NORTHERN |OTHER OFFENSES|172231.0|1.0 |2.0 |(7,[1],[1.0])|(10,[2],[1.0])|(389257,[172231],[1.0])|1.0 |(389274,[172231,389258,389266],[1.0,1.0,1.0])| |2015-05-13 18:05:00|Wednesday|BAYVIEW |LARCENY/THEFT |330092.0|1.0 |3.0 |(7,[1],[1.0])|(10,[3],[1.0])|(389257,[330092],[1.0])|0.0 |(389274,[330092,389258,389267],[1.0,1.0,1.0])| |2015-05-13 18:02:00|Wednesday|MISSION |OTHER OFFENSES|387792.0|1.0 |1.0 |(7,[1],[1.0])|(10,[1],[1.0])|(389257,[387792],[1.0])|1.0 |(389274,[387792,389258,389265],[1.0,1.0,1.0])| |2015-05-13 18:00:00|Wednesday|SOUTHERN |BURGLARY |32607.0 |1.0 |0.0 |(7,[1],[1.0])|(10,[0],[1.0])|(389257,[32607],[1.0]) |8.0 |(389274,[32607,389258,389264],[1.0,1.0,1.0]) | |2015-05-13 18:00:00|Wednesday|BAYVIEW |LARCENY/THEFT |32607.0 |1.0 |3.0 |(7,[1],[1.0])|(10,[3],[1.0])|(389257,[32607],[1.0]) |0.0 |(389274,[32607,389258,389267],[1.0,1.0,1.0]) | |2015-05-13 18:00:00|Wednesday|PARK |LARCENY/THEFT |32607.0 |1.0 |8.0 |(7,[1],[1.0])|(10,[8],[1.0])|(389257,[32607],[1.0]) |0.0 |(389274,[32607,389258,389272],[1.0,1.0,1.0]) | +-------------------+---------+----------+--------------+--------+-----+-----+-------------+--------------+-----------------------+-----+---------------------------------------------+ only showing top 7 rows
******************************************************************************************************************* 数据2： id,name,age,sex,rate 1,lyy,20,F,0.6 2,rdd,20,M,0.4 3,nyc,18,M,0.55 4,mzy,10,M,0.21
1 //Binarizer二值化： 将该列数据二值化，大于阈值的为1.0，否则为0.0 spark源码:udf { in: Double => if (in > td) 1.0 else 0.0 } 2 3 Dataset<Row> result = new Binarizer() 4 .setInputCol("rate") 5 .setOutputCol("flag") 6 .setThreshold(0.5).transform(data); 7 8 result.show(10, false);
+---+----+---+---+----+----+ |id |name|age|sex|rate|flag| +---+----+---+---+----+----+ |1 |lyy |20 |F |0.6 |1.0 | |2 |rdd |20 |M |0.4 |0.0 | |3 |nyc |18 |M |0.55|1.0 | |4 |mzy |10 |M |0.21|0.0 | +---+----+---+---+----+----+
1 //IndexToString将stringindexder转换的数据转回到原始的数据 2 3 StringIndexer labelIndexer = new StringIndexer() 4 .setInputCol("sex") 5 .setOutputCol("label"); 6 7 IndexToString IndexToSex = new IndexToString() 8 .setInputCol("label") 9 .setOutputCol("orisex"); 10 11 Pipeline pipeline = new Pipeline().setStages(new PipelineStage[] { labelIndexer, IndexToSex}); 12 PipelineModel model = pipeline.fit(data); 13 14 // Make predictions. 15 Dataset<Row> result = model.transform(data); 16 17 result.show(10, false);
1 //Bucketizer 分箱（分段处理）：将连续数值转换为离散类别 2 //比如特征是年龄，是一个连续数值，需要将其转换为离散类别(未成年人、青年人、中年人、老年人），就要用到Bucketizer了 3 //如age > 55 老年人 4 double[] splits={0,18,35,55,Double.POSITIVE_INFINITY};//[0,18),[18,35),[35,55),[55,正无穷) 5 Dataset<Row> result=new Bucketizer() 6 .setInputCol("age") 7 .setOutputCol("bucketCategory") 8 .setSplits(splits)//设置分段标准 9 .transform(data); 10 11 result.show(10, false);
