问题:

Now I have 300+ columns in my RDD, but I found there is a need to dynamically select a range of columns and put them into LabledPoints data type. As a newbie to Spark, I am wondering if there is any index way to select a range of columns in RDD. Something like temp_data = data[, 101:211] in R. Is there something like val temp_data = data.filter(_.column_index in range(101:211)...?

Any thought is welcomed and appreciated.

回答1:

If it is a DataFrame, then something like this should work:

val df = rdd.toDF
df.select(df.columns.slice(101,211) : _*)

回答2:

Assuming you have an RDD of Array or any other scala collection (e.g., List). You can do something like this:

val data: RDD[Array[Int]] = sc.parallelize(Array(Array(1,2,3), Array(4,5,6)))
val sliced: RDD[Array[Int]] = data.map(_.slice(0,2))

sliced.collect()
> Array[Array[Int]] = Array(Array(1, 2), Array(4, 5))

回答3:

Kind of old thread, but I recently had to do something similar and search around. I needed to select all but the last column where I had 200+ columns.

Spark 1.4.1
Scala 2.10.4

val df = hiveContext.sql("SELECT * FROM foobar")
val cols = df.columns.slice(0, df.columns.length - 1)
val new_df = df.select(cols.head, cols.tail:_*)