pyspark change day in datetime column

问题

what is wrong with this code trying to change day of a datetime columns

import pyspark
import pyspark.sql.functions as sf
import pyspark.sql.types as sparktypes
import datetime

sc = pyspark.SparkContext(appName="test")
sqlcontext = pyspark.SQLContext(sc)

rdd = sc.parallelize([('a',datetime.datetime(2014, 1, 9, 0, 0)),
                      ('b',datetime.datetime(2014, 1, 27, 0, 0)),
                      ('c',datetime.datetime(2014, 1, 31, 0, 0))])
testdf = sqlcontext.createDataFrame(rdd, ["id", "date"])

print(testdf.show())
print(testdf.printSchema())

gives a test dataframe:

+---+--------------------+
| id|                date|
+---+--------------------+
|  a|2014-01-09 00:00:...|
|  b|2014-01-27 00:00:...|
|  c|2014-01-31 00:00:...|
+---+--------------------+


root
 |-- id: string (nullable = true)
 |-- date: timestamp (nullable = true)

Then I define a udf to change day of date column:

def change_day_(date, day):
    return date.replace(day=day)

change_day = sf.udf(change_day_, sparktypes.TimestampType())
testdf.withColumn("PaidMonth", change_day(testdf.date, 1)).show(1)

This raises an error:

Py4JError: An error occurred while calling z:org.apache.spark.sql.functions.col. Trace:
py4j.Py4JException: Method col([class java.lang.Integer]) does not exist
    at py4j.reflection.ReflectionEngine.getMethod(ReflectionEngine.java:318)
    at py4j.reflection.ReflectionEngine.getMethod(ReflectionEngine.java:339)
    at py4j.Gateway.invoke(Gateway.java:274)
    at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:132)
    at py4j.commands.CallCommand.execute(CallCommand.java:79)
    at py4j.GatewayConnection.run(GatewayConnection.java:214)
    at java.lang.Thread.run(Thread.java:745)

回答1:

A udf which recieves multiple arguments is assumed to recieve multiple columns. The "1" is not a column.

This means you can do one of the following. Either make it a column as suggested in the comments:

testdf.withColumn("PaidMonth", change_day(testdf.date, lit(1))).show(1)

lit(1) is a column of ones

or make the original function return a higher order function:

def change_day_(day):
    return lambda date: date.replace(day=day)

change_day = sf.udf(change_day_(1), sparktypes.TimestampType())
testdf.withColumn("PaidMonth", change_day(testdf.date)).show(1)

This basically creates a function which replaces with 1 and therefore can recieve an integer. The udf would apply on a single column.

回答2:

Thanks to @ArthurTacca's comment, the trick is to use pyspark.sql.functions.lit() function like this:

testdf.withColumn("PaidMonth", change_day(testdf.date, sf.lit(1))).show()

alternate answers welcome!

来源：https://stackoverflow.com/questions/42583433/pyspark-change-day-in-datetime-column