Summing n columns in Spark in Java using dataframes

问题

String[] col = {"a","b","c"}

Data:

id a b c d e 
101 1 1 1 1 1
102 2 2 2 2 2
103 3 3 3 3 3

Expected output:- id with sum of columns specified in column string

id (a+b+c)
101 3
102 6
103 9

How to do this using dataframes?

回答1:

if you are using java you can do the following

import org.apache.spark.SparkConf;
import org.apache.spark.SparkContext;
import org.apache.spark.sql.Dataset;
import org.apache.spark.sql.Row;
import org.apache.spark.sql.SQLContext;
import org.apache.spark.sql.types.DataTypes;

static SparkConf conf = new SparkConf().setMaster("local").setAppName("simple");
static SparkContext sc = new SparkContext(conf);
static SQLContext sqlContext = new SQLContext(sc);

public static void main(String[] args) {

    Dataset<Row> df = sqlContext.read()
            .format("com.databricks.spark.csv")
            .option("delimiter", " ")
            .option("header", true)
            .option("inferSchema", true)
            .load("path to the input text file");


    sqlContext.udf().register("sums", (Integer a, Integer b, Integer c) -> a+b+c, DataTypes.IntegerType);
    df.registerTempTable("temp");
    sqlContext.sql("SELECT id, sums(a, b, c) AS `(a+b+c)` FROM temp").show(false);

}

and you should have output as

+---+-------+
|id |(a+b+c)|
+---+-------+
|101|3      |
|102|6      |
|103|9      |
+---+-------+

If you prefer to go without sql query and use api then you can do as below

import org.apache.spark.sql.expressions.UserDefinedFunction;
import org.apache.spark.sql.types.DataTypes;
import static org.apache.spark.sql.functions.col;
import static org.apache.spark.sql.functions.udf;

    UserDefinedFunction mode = udf((Integer a, Integer b, Integer c) -> a+b+c, DataTypes.IntegerType);
    df.select(col("id"), mode.apply(col("a"), col("b"), col("c")).as("(a+b+c)")).show(false);

回答2:

You can create a string with the expression and then use expr to create the column. In other words, in this case you want to create the string "a+b+c" which you then can use. This will work for any number of columns.

In Scala it can look as follows (it should be fairly simple to translate to Java):

import org.apache.spark.sql.functions.expr

val df = Seq((101,1,1,1,1,1),(102,2,2,2,2,2),(103,3,3,3,3,3)).toDF("id", "a", "b", "c", "d", "e") 

val cols = Seq("a", "b", "c")
val expression = cols.mkString("+")
val colName = "(" + expression + ")"
df.select($"id", expr(expression).as(colName))

which will give you:

+---+-------+
| id|(a+b+c)|
+---+-------+
|101|      3|
|102|      6|
|103|      9|
+---+-------+

回答3:

There are many different ways to do this. You might use a map, like this:

val df = Seq((101,1,1,1,1,1),(102,2,2,2,2,2),(103,3,3,3,3,3)).toDF("id", "a", "b", "c", "d", "e")

df.map(row => (row.getString(0), row.getInt(1)+row.getInt(2)+row.getInt(3)))
  .toDF("id", "a+b+c")

Or you could use a udf, like this:

import org.apache.spark.sql.functions._
import spark.implicits._

val addCols = udf((a: Int, b:Int, c: Int) => a+b+c)
df.select('id, addCols('a, 'b, 'c) as "a+b+c")

Or go with Shaido's suggestion :)

回答4:

This works for me in Java:

final var allDataFamilyDf = allDataDf.withColumn("FamilySize",
    functions.col("SibSp").plus(functions.col("Parch")));

回答5:

A more cleaner Java way of doing this (as mentioned by @shaido-reinstate-monica):

String[] columnNames = {"a","b","c"};      // columnNames is the list of column names to be added together
Buffer<Column> sums = JavaConversions.asScalaBuffer(ImmutableList.of(columnNames).stream().map(name -> col(name)).collect(Collectors.toList()));

String expression = sums.mkString("+");
df.selectExpr("id", expression);     // where df is the dataset with columns "id", "a", "b", and "c"

来源：https://stackoverflow.com/questions/50248228/summing-n-columns-in-spark-in-java-using-dataframes

标签

java

apache-spark

apache-spark-sql

spark-dataframe