如何在 pyspark 中将字符串数组转换为日期？

Question

This is the dataframe schema structure.

root
 |-- validFrom: array (nullable = true)
 |    |-- element: string (containsNull = true)

This is the pyspark query where I try to get the date datatype for validFrom .

df_02 = spark.sql("""
 select to_date(validFrom, 'yyyy-MM-dd') as  validFrom
 from   v_source_df_flatten
""");

But I receive the following error.

due to data type mismatch: argument 1 requires (string or date or timestamp) type, however, 'v_starhist_df_flatten. validFrom ' is of array type.

What do I have to change in the to_date function?

Answer 1

You will have to apply to_date on each element of the array using transform .

query = """
SELECT transform(validFrom, x -> to_date(x, 'yyyy-MM-dd')) as validFrom
FROM v_source_df_flatten
"""

df_02 = spark.sql(query)

df_02.printSchema()

"""
root
 |-- validFrom: array (nullable = true)
 |    |-- element: date (containsNull = true)
"""

df_02.show(truncate=False)

"""
+------------------------+
|validFrom               |
+------------------------+
|[2021-10-10, 2022-10-10]|
|[2021-01-01, 2022-01-01]|
+------------------------+
"""

Answer 2

You can't convert an array of string directly into DateType . to_date function expects a string date.

If you have only one date per array, then you can access simply the first element of the array and convert it to date like this:

spark.createDataFrame(
    [(["2022-01-01"],), (["2022-01-02"],)], ["validFrom"]
).createOrReplaceTempView("v_source_df_flatten")

df_02 = spark.sql("select to_date(validFrom[0]) as validFrom from v_source_df_flatten")

df_02.printSchema()
#root
# |-- validFrom: date (nullable = true)

df_02.show()
#+----------+
#| validFrom|
#+----------+
#|2022-01-01|
#|2022-01-02|
#+----------+

Notice that you can also use a simple cast as your dates have the default pattern yyyy-MM-dd :

cast(validFrom[0] as date) as validFrom

However, if your intent is to convert an array of strings into an array of dates, then you can use a cast in this particular case:

df_02 = spark.sql("select cast(validFrom as array<date>) as validFrom from v_source_df_flatten")

df_02.printSchema()
#root
# |-- validFrom: array (nullable = true)
# |    |-- element: date (containsNull = true)

df_02.show()
#+------------+
#|   validFrom|
#+------------+
#|[2022-01-01]|
#|[2022-01-02]|
#+------------+

For date pattern different from yyyy-MM-dd you'll have to use transform on the array and apply to_date for each element as shown in the other answer.