Group spark dataframe by date

Question

I've loaded a DataFrame from a SQLServer table. It looks like this:

>>> df.show()
+--------------------+----------+
|           timestamp|    Value |
+--------------------+----------+
|2015-12-02 00:10:...|     652.8|
|2015-12-02 00:20:...|     518.4|
|2015-12-02 00:30:...|     524.6|
|2015-12-02 00:40:...|     382.9|
|2015-12-02 00:50:...|     461.6|
|2015-12-02 01:00:...|     476.6|
|2015-12-02 01:10:...|     472.6|
|2015-12-02 01:20:...|     353.0|
|2015-12-02 01:30:...|     407.9|
|2015-12-02 01:40:...|     475.9|
|2015-12-02 01:50:...|     513.2|
|2015-12-02 02:00:...|     569.0|
|2015-12-02 02:10:...|     711.4|
|2015-12-02 02:20:...|     457.6|
|2015-12-02 02:30:...|     392.0|
|2015-12-02 02:40:...|     459.5|
|2015-12-02 02:50:...|     560.2|
|2015-12-02 03:00:...|     252.9|
|2015-12-02 03:10:...|     228.7|
|2015-12-02 03:20:...|     312.2|
+--------------------+----------+

Now I'd like to group (and sum) values by hour (or day, or month or...), but I don't really have a clue about how can I do that.

That's how I load the DataFrame. I've got the feeling that this isn't the right way to do it, though:

query = """
SELECT column1 AS timestamp, column2 AS value
FROM table
WHERE  blahblah
"""

sc = SparkContext("local", 'test')
sqlctx = SQLContext(sc)

df = sqlctx.load(source="jdbc",
                 url="jdbc:sqlserver://<CONNECTION_DATA>",
                 dbtable="(%s) AS alias" % query)

Is it ok?

Answer 1

Since 1.5.0 Spark provides a number of functions like dayofmonth, hour, month or year which can operate on dates and timestamps. So if timestamp is a TimestampType all you need is a correct expression. For example:

from pyspark.sql.functions import hour, mean

(df
    .groupBy(hour("timestamp").alias("hour"))
    .agg(mean("value").alias("mean"))
    .show())

## +----+------------------+
## |hour|              mean|
## +----+------------------+
## |   0|508.05999999999995|
## |   1| 449.8666666666666|
## |   2| 524.9499999999999|
## |   3|264.59999999999997|
## +----+------------------+

Pre-1.5.0 your best option is to use HiveContext and Hive UDFs either with selectExpr:

df.selectExpr("year(timestamp) AS year", "value").groupBy("year").sum()

## +----+---------+----------+   
## |year|SUM(year)|SUM(value)|
## +----+---------+----------+
## |2015|    40300|    9183.0|
## +----+---------+----------+

or raw SQL:

df.registerTempTable("df")

sqlContext.sql("""
    SELECT MONTH(timestamp) AS month, SUM(value) AS values_sum
    FROM df
    GROUP BY MONTH(timestamp)""")

Just remember that aggregation is performed by Spark not pushed-down to the external source. Usually it is a desired behavior but there are situations when you may prefer to perform aggregation as a subquery to limit data transfer.

Answer 2

Also, you can use date_format to create any time period you wish. Groupby specific day:

from pyspark.sql import functions as F

df.select(F.date_format('timestamp','yyyy-MM-dd').alias('day')).groupby('day').count().show()

Groupby specific month (just change the format):

df.select(F.date_format('timestamp','yyyy-MM').alias('month')).groupby('month').count().show()