I'm trying to add an "CASE WHEN ... ELSE ..." calculated column to an existing DataFrame, using Scala APIs.
Starting dataframe:
color
Red
Green
Blue
Desired dataframe (SQL syntax: CASE WHEN color == Green THEN 1 ELSE 0 END AS bool):
color bool
Red 0
Green 1
Blue 0
How should I implement this logic?
In the upcoming SPARK 1.4.0 release (should be released in the next couple of days). You can use the when/otherwise syntax:
// Create the dataframe
val df = Seq("Red", "Green", "Blue").map(Tuple1.apply).toDF("color")
// Use when/otherwise syntax
val df1 = df.withColumn("Green_Ind", when($"color" === "Green", 1).otherwise(0))
If you are using SPARK 1.3.0 you can chose to use a UDF:
// Define the UDF
val isGreen = udf((color: String) => {
if (color == "Green") 1
else 0
})
val df2 = df.withColumn("Green_Ind", isGreen($"color"))
In Spark 1.5.0: you can also use the SQL syntax expr function
val df3 = df.withColumn("Green_Ind", expr("case when color = 'green' then 1 else 0 end"))
or plain spark-sql
df.registerTempTable("data")
val df4 = sql(""" select *, case when color = 'green' then 1 else 0 end as Green_ind from data """)
I found this:
https://issues.apache.org/jira/browse/SPARK-3813
Worked for me on spark 2.1.0:
import sqlContext._
val rdd = sc.parallelize((1 to 100).map(i => Record(i, s"val_$i")))
rdd.registerTempTable("records")
println("Result of SELECT *:")
sql("SELECT case key when '93' then 'ravi' else key end FROM records").collect()
I was looking for that long time so here is example of SPARK 2.1 JAVA with group by- for other java users.
import static org.apache.spark.sql.functions.*;
//...
Column uniqTrue = col("uniq").equalTo(true);
Column uniqFalse = col("uniq").equalTo(false);
Column testModeFalse = col("testMode").equalTo(false);
Column testModeTrue = col("testMode").equalTo(true);
Dataset<Row> x = basicEventDataset
.groupBy(col(group_field))
.agg(
sum(when((testModeTrue).and(uniqTrue), 1).otherwise(0)).as("tt"),
sum(when((testModeFalse).and(uniqTrue), 1).otherwise(0)).as("ft"),
sum(when((testModeTrue).and(uniqFalse), 1).otherwise(0)).as("tf"),
sum(when((testModeFalse).and(uniqFalse), 1).otherwise(0)).as("ff")
);