Spark DataFrame不尊重模式并将所有内容都视为String

Question

我正面临着一个问题,我现在已经很久没能克服.

1. 我在Spark 1.4和Scala 2.10上.此时我无法升级(大型分布式基础架构)

2. 我有一个包含几百列的文件,其中只有两列是字符串,其余都是Long.我想将此数据转换为标签/功能数据帧.

3. 我已经能够将它变成LibSVM格式.

4. 我只是无法将其变成标签/功能格式.

原因是

1. 我无法使用此处显示的toDF() https://spark.apache.org/docs/1.5.1/ml-ensembles.html

   val data = MLUtils.loadLibSVMFile(sc, "data/mllib/sample_libsvm_data.txt").toDF()
   

   它在1.4中不受支持

2. 所以我首先将txtFile转换为DataFrame,我使用了这样的东西

   def getColumnDType(columnName:String):StructField = {
   
           if((columnName== "strcol1") || (columnName== "strcol2")) 
               return StructField(columnName, StringType, false)
           else
               return StructField(columnName, LongType, false)
       }
   def getDataFrameFromTxtFile(sc: SparkContext,staticfeatures_filepath: String,schemaConf: String) : DataFrame = {
           val sfRDD = sc.textFile(staticfeatures_filepath)//
           val sqlContext = new org.apache.spark.sql.SQLContext(sc)
            // reads a space delimited string from application.properties file
           val schemaString = readConf(Array(schemaConf)).get(schemaConf).getOrElse("")
   
           // Generate the schema based on the string of schema
           val schema =
             StructType(
               schemaString.split(" ").map(fieldName => getSFColumnDType(fieldName)))
   
           val data = sfRDD
           .map(line => line.split(","))
           .map(p => Row.fromSeq(p.toSeq))
   
           var df = sqlContext.createDataFrame(data, schema)
   
           //schemaString.split(" ").drop(4)
           //.map(s => df = convertColumn(df, s, "int"))
   
           return df
       }   
   

当我df.na.drop() df.printSchema()使用这个返回的数据帧时,我会得到完美的Schema

root
 |-- rand_entry: long (nullable = false)
 |-- strcol1: string (nullable = false)
 |-- label: long (nullable = false)
 |-- strcol2: string (nullable = false)
 |-- f1: long (nullable = false)
 |-- f2: long (nullable = false)
 |-- f3: long (nullable = false)
and so on till around f300

但是 - 可悲的部分是我尝试用df做的任何事情(见下文),我总是得到一个ClassCastException(java.lang.String不能强制转换为java.lang.Long)

val featureColumns = Array("f1","f2",....."f300")
assertEquals(-99,df.select("f1").head().getLong(0))
assertEquals(-99,df.first().get(4))
val transformeddf = new VectorAssembler()
        .setInputCols(featureColumns)
        .setOutputCol("features")
        .transform(df)

所以 - 糟糕的是 - 即使模式显示为Long - df仍然在内部将所有内容都视为字符串.

编辑

添加一个简单的例子

说我有这样的文件

1,A,20,P,-99,1,0,0,8,1,1,1,1,131153
1,B,23,P,-99,0,1,0,7,1,1,0,1,65543
1,C,24,P,-99,0,1,0,9,1,1,1,1,262149
1,D,7,P,-99,0,0,0,8,1,1,1,1,458759

和

sf-schema=f0 strCol1 f1 strCol2 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11

(列名真的无所谓所以你可以忽略这些细节)

我所要做的就是创建一个Label/Features类型的数据框,其中我的第3列成为标签,第5到第11列成为要素[Vector]列.这样我就可以按照https://spark.apache.org/docs/latest/ml-classification-regression.html#tree-ensembles中的其他步骤进行操作.

我已按照零323的建议抛出列

val types = Map("strCol1" -> "string", "strCol2" -> "string")
        .withDefault(_ => "bigint")
df = df.select(df.columns.map(c => df.col(c).cast(types(c)).alias(c)): _*)
df = df.drop("f0")
df = df.drop("strCol1")
df = df.drop("strCol2")

但断言和VectorAssembler仍然失败.featureColumns = Array("f2","f3",....."f11")这是我拥有df之后想做的整个序列

    var transformeddf = new VectorAssembler()
    .setInputCols(featureColumns)
    .setOutputCol("features")
    .transform(df)

    transformeddf.show(2)

    transformeddf = new StringIndexer()
    .setInputCol("f1")
    .setOutputCol("indexedF1")
    .fit(transformeddf)
    .transform(transformeddf)

    transformeddf.show(2)

    transformeddf = new VectorIndexer()
    .setInputCol("features")
    .setOutputCol("indexedFeatures")
    .setMaxCategories(5)
    .fit(transformeddf)
    .transform(transformeddf)

来自ScalaIDE的异常跟踪 - 就在它击中VectorAssembler时如下所示

java.lang.ClassCastException: java.lang.String cannot be cast to java.lang.Long
    at scala.runtime.BoxesRunTime.unboxToLong(BoxesRunTime.java:110)
    at scala.math.Numeric$LongIsIntegral$.toDouble(Numeric.scala:117)
    at org.apache.spark.sql.catalyst.expressions.Cast$$anonfun$castToDouble$5.apply(Cast.scala:364)
    at org.apache.spark.sql.catalyst.expressions.Cast$$anonfun$castToDouble$5.apply(Cast.scala:364)
    at org.apache.spark.sql.catalyst.expressions.Cast.eval(Cast.scala:436)
    at org.apache.spark.sql.catalyst.expressions.Alias.eval(namedExpressions.scala:118)
    at org.apache.spark.sql.catalyst.expressions.CreateStruct$$anonfun$eval$2.apply(complexTypes.scala:75)
    at org.apache.spark.sql.catalyst.expressions.CreateStruct$$anonfun$eval$2.apply(complexTypes.scala:75)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:244)
    at scala.collection.mutable.ResizableArray$class.foreach(ResizableArray.scala:59)
    at scala.collection.mutable.ArrayBuffer.foreach(ArrayBuffer.scala:47)
    at scala.collection.TraversableLike$class.map(TraversableLike.scala:244)
    at scala.collection.AbstractTraversable.map(Traversable.scala:105)
    at org.apache.spark.sql.catalyst.expressions.CreateStruct.eval(complexTypes.scala:75)
    at org.apache.spark.sql.catalyst.expressions.CreateStruct.eval(complexTypes.scala:56)
    at org.apache.spark.sql.catalyst.expressions.ScalaUdf$$anonfun$2.apply(ScalaUdf.scala:72)
    at org.apache.spark.sql.catalyst.expressions.ScalaUdf$$anonfun$2.apply(ScalaUdf.scala:70)
    at org.apache.spark.sql.catalyst.expressions.ScalaUdf.eval(ScalaUdf.scala:960)
    at org.apache.spark.sql.catalyst.expressions.Alias.eval(namedExpressions.scala:118)
    at org.apache.spark.sql.catalyst.expressions.InterpretedMutableProjection.apply(Projection.scala:68)
    at org.apache.spark.sql.catalyst.expressions.InterpretedMutableProjection.apply(Projection.scala:52)
    at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
    at scala.collection.Iterator$$anon$11.next(Iterator.scala:328)
    at scala.collection.Iterator$$anon$10.next(Iterator.scala:312)
    at scala.collection.Iterator$class.foreach(Iterator.scala:727)
    at scala.collection.AbstractIterator.foreach(Iterator.scala:1157)
    at scala.collection.generic.Growable$class.$plus$plus$eq(Growable.scala:48)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:103)
    at scala.collection.mutable.ArrayBuffer.$plus$plus$eq(ArrayBuffer.scala:47)
    at scala.collection.TraversableOnce$class.to(TraversableOnce.scala:273)
    at scala.collection.AbstractIterator.to(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toBuffer(TraversableOnce.scala:265)
    at scala.collection.AbstractIterator.toBuffer(Iterator.scala:1157)
    at scala.collection.TraversableOnce$class.toArray(TraversableOnce.scala:252)
    at scala.collection.AbstractIterator.toArray(Iterator.scala:1157)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$3.apply(SparkPlan.scala:143)
    at org.apache.spark.sql.execution.SparkPlan$$anonfun$3.apply(SparkPlan.scala:143)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1767)
    at org.apache.spark.SparkContext$$anonfun$runJob$5.apply(SparkContext.scala:1767)
    at org.apache.spark.scheduler.ResultTask.runTask(ResultTask.scala:63)
    at org.apache.spark.scheduler.Task.run(Task.scala:70)
    at org.apache.spark.executor.Executor$TaskRunner.run(Executor.scala:213)
    at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1145)
    at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:615)
    at java.lang.Thread.run(Thread.java:745)

Answer 1

你得到ClassCastException因为这正是应该发生的事情.Schema参数不用于自动转换(有些DataSources可能以这种方式使用模式,但不能使用类似的方法createDataFrame).它只声明存储在行中的值的类型.您有责任传递与架构匹配的数据,而不是相反.

虽然DataFrame显示模式,但您已声明它仅在运行时验证,因此运行时异常.如果要将数据转换为特定cast数据,则显式拥有数据.

1. 首先阅读所有列StringType:

   val rows = sc.textFile(staticfeatures_filepath)
     .map(line => Row.fromSeq(line.split(",")))
   
   val schema = StructType(
     schemaString.split(" ").map(
       columnName => StructField(columnName, StringType, false)
     )
   )
   
   val df = sqlContext.createDataFrame(rows, schema)
   

2. 接下来将所选列转换为所需类型:

   import org.apache.spark.sql.types.{LongType, StringType}
   
   val types = Map("strcol1" -> StringType, "strcol2" -> StringType)
     .withDefault(_ => LongType)
   
   val casted = df.select(df.columns.map(c => col(c).cast(types(c)).alias(c)): _*)
   

3. 使用汇编程序:

   val transformeddf = new VectorAssembler()
     .setInputCols(featureColumns)
     .setOutputCol("features")
     .transform(casted)
   

您只需使用spark-csv以下步骤1和2 :

// As originally 
val schema = StructType(
  schemaString.split(" ").map(fieldName => getSFColumnDType(fieldName)))


val df = sqlContext
  .read.schema(schema)
  .format("com.databricks.spark.csv")
  .option("header", "false")
  .load(staticfeatures_filepath)

另请参阅在PySpark中正确读取文件中的类型