问题
From a dataframe I want to get names of columns which contain at least one null value inside.
Considering the dataframe below:
val dataset = sparkSession.createDataFrame(Seq(
(7, null, 18, 1.0),
(8, "CA", null, 0.0),
(9, "NZ", 15, 0.0)
)).toDF("id", "country", "hour", "clicked")
I want to get column names 'Country' and 'Hour'.
id country hour clicked
7 null 18 1
8 "CA" null 0
9 "NZ" 15 0
回答1:
this is one solution, but it's a bit awkward, I hope there is an easier way:
val cols = dataset.columns
val columnsToSelect = dataset
// count null values (by summing up 1s if its null)
.select(cols.map(c => (sum(when(col(c).isNull,1))>0).alias(c)):_*)
.head() // collect result of aggregation
.getValuesMap[Boolean](cols) // now get columns which are "true"
.filter{case (c,hasNulls) => hasNulls}
.keys.toSeq // and get the name of those columns
dataset
.select(columnsToSelect.head,columnsToSelect.tail:_*)
.show()
+-------+----+
|country|hour|
+-------+----+
| null| 18|
| CA|null|
| NZ| 15|
+-------+----+
回答2:
A slight modification of this answer, comparing the counts per column to the number of rows:
import org.apache.spark.sql.functions.{count,col}
// Get number of rows
val nr_rows = dataset.count
// Get column indices
val col_inds = dataset.select(dataset.columns.map(c => count(col(c)).alias(c)): _*)
.collect()(0)
.toSeq.zipWithIndex
.filter(_._1 != nr_rows).map(_._2)
// Subset column names using the indices
col_inds.map(i => dataset.columns.apply(i))
Seq[String] = ArrayBuffer(country, hour)
来源:https://stackoverflow.com/questions/48261746/spark-get-only-columns-that-have-one-or-more-null-values