我不能使用scala在apache spark中使用流模式进行在线预测的数据帧

Question

I am new in spark and I would like to make a streaming program. I need to predict a number repetition for each of my rows. Here is my raw data:

05:49:56.604899 00:00:00:00:00:02 > 00:00:00:00:00:03, ethertype IPv4 (0x0800), length 10202: 10.0.0.2.54880 > 10.0.0.3.5001: Flags [.], seq 3641977583:3641987719, ack 129899328, win 58, options [nop,nop,TS val 432623 ecr 432619], length 10136
05:49:56.604908 00:00:00:00:00:03 > 00:00:00:00:00:02, ethertype IPv4 (0x0800), length 66: 10.0.0.3.5001 > 10.0.0.2.54880: Flags [.], ack 10136, win 153, options [nop,nop,TS val 432623 ecr 432623], length 0
05:49:56.604900 00:00:00:00:00:02 > 00:00:00:00:00:03, ethertype IPv4 (0x0800), length 4410: 10.0.0.2.54880 > 10.0.0.3.5001: Flags [P.], seq 10136:14480, ack 1, win 58, options [nop,nop,TS val 432623 ecr 432619], length 4344

I wrote a code that extract my suitable output like below. (I needed the number of repetition on column1 and column2)

Here is my code:

However my code is not in a streaming mode. I did another code to obtain a streaming mode. Because the train.csv file is generating in streaming way. But I got some errors. Here is my streaming code:

import org.apache.spark.SparkConf
import org.apache.spark.mllib.linalg.Vectors
import org.apache.spark.mllib.regression.{LabeledPoint, StreamingLinearRegressionWithSGD}
import org.apache.spark.sql.Row
import org.apache.spark.sql.types.{StringType, StructField, StructType}
import org.apache.spark.streaming.{Seconds, StreamingContext}

import scala.util.Try
/**
  * Created by saeedtkh on 5/24/17.
  */
object Main_ML_with_Streaming {
  def main(args: Array[String]) {

    val conf = new SparkConf().setAppName("saeed_test").setMaster("local[*]")
    //val sc = new SparkContext(conf)
    val ssc = new StreamingContext(conf, Seconds(5))


    /////////////////////Start extract the packet
    val customSchema = StructType(Array(
      StructField("column0", StringType, true),
      StructField("column1", StringType, true),
      StructField("column2", StringType, true)))


    val rdd = ssc.textFileStream("/Users/saeedtkh/Desktop/sharedsaeed/train.csv")
    val rowRdd =rdd.map(line => line.split(">")).map(array => {
      val first = Try(array(0).trim.split(" ")(0)) getOrElse ""
      val second = Try(array(1).trim.split(" ")(6)) getOrElse ""
      val third = Try(array(2).trim.split(" ")(0).replace(":", "")) getOrElse ""
      Row.fromSeq(Seq(first, second, third))
    })

    val dataFrame_trainingData = sqlContext.createDataFrame(rowRdd, customSchema)
    dataFrame_trainingData.groupBy("column1","column2").count().show()

    /////////////////////end extract the packet

    val testData = ssc.textFileStream(/Users/saeedtkh/Desktop/sharedsaeed/test.csv).map(LabeledPoint.parse)
    ////////////////////end trainging and testing

    val numFeatures = 3
    val model = new StreamingLinearRegressionWithSGD()
      .setInitialWeights(Vectors.zeros(numFeatures))

    model.trainOn(dataFrame_trainingData)
    model.predictOnValues(testData.map(lp => (lp.label, lp.features))).print()

    ssc.start()
    ssc.awaitTermination()

    print("Here is the anwser: *****########*********#########*******222")
  }
}

The problem is that, I can not create a dataframe using sqlcontext in this line in my code:

val dataFrame_trainingData = sqlContext.createDataFrame(rowRdd, customSchema)

Can any body help me to modify this code that works in streaming way and predicts repetition of each row using Linear Regression or any other algorithm. Thanks a lot.

Update1: Acoording to answer number one, I added foreach but errors are still exists:

Answer 1

First, it's important to note that ssc.textFileStream returns a DStream and not an RDD , so the variables you named rdd , rowRdd and testData are not really RDDs, but rather abstractions over a continuous sequence of RDDs. Therefore, you cannot pass these to createDataFrame which expects RDDs.

You can create a DataFrame out of each underlying RDD using DStream.foreachRDD , as described here :

rowRdd.foreachRDD { rdd =>
  val dataFrame_trainingData = sqlContext.createDataFrame(rdd, customSchema)
  // ...
} 

However, you should notice that StreamingLinearRegressionWithSGD expects DStreams as inputs for trainOn and predictOnValues - so you can simply pass the original DStreams without converting them into DataFrames.