`Найден обходной путь, посмотрите, есть ли лучшее решение с использованием одного кадра данных и без UDF '
import org.apache.spark.sql.functions.regexp_replace
val df = spark.createDataFrame(Seq(
(1, "9/11/2020"),
(2, "10/11/2020"),
(3, "1/1/2020"),
(4, "12/7/2020"))).toDF("Id", "x4")
val newDf = df.withColumn("x4New", regexp_replace(df("x4"), "(?:(\\b\\d{2}))/(?:(\\d))/(?:(\\d{4})\\b)", "$1/0$2/$3"))
val newDf1 = newDf.withColumn("x4New1", regexp_replace(newDf("x4New"), "(?:(\\b\\d{1}))/(?:(\\d))/(?:(\\d{4})\\b)", "0$1/$2/$3"))
val newDf2 = newDf1.withColumn("x4New2", regexp_replace(newDf1("x4New1"), "(?:(\\b\\d{1}))/(?:(\\d{2}))/(?:(\\d{4})\\b)", "0$1/$2/$3"))
val newDf3 = newDf2.withColumn("date", to_date(regexp_replace(newDf2("x4New2"), "(?:(\\b\\d{2}))/(?:(\\d{1}))/(?:(\\d{4})\\b)", "$1/0$2/$3"),"MM/dd/yyyy"))
val formatedDataDf = newDf3
.drop("x4New")
.drop("x4New1")
.drop("x4New2")
formatedDataDf.printSchema
formatedDataDf.show
Output looks like as follows
root
|-- Id: integer (nullable = false)
|-- x4: string (nullable = true)
|-- date: date (nullable = true)
+---+----------+----------+
| Id| x4| date|
+---+----------+----------+
| 1| 9/11/2020|2020-09-11|
| 2|10/11/2020|2020-10-11|
| 3| 1/1/2020|2020-01-01|
| 4| 12/7/2020|2020-12-07|
+---+----------+----------+