我有一个如下所示的 Spark DataFrame:
+---------+--------------------------+
|group_id |event_time |
+---------+--------------------------+
|XXXX |2017-10-25 14:47:02.717013|
|XXXX |2017-10-25 14:47:25.444979|
|XXXX |2017-10-25 14:49:32.21353 |
|YYYY |2017-10-25 14:50:38.321134|
|YYYY |2017-10-25 14:51:12.028447|
|ZZZZ |2017-10-25 14:51:24.810688|
|YYYY |2017-10-25 14:37:34.241097|
|ZZZZ |2017-10-25 14:37:24.427836|
|XXXX |2017-10-25 14:37:24.620864|
|YYYY |2017-10-25 14:37:24.964614|
+---------+--------------------------+
我想计算每个 group_id 一天内每小时事件的滚动计数。
因此，对于日期时间 25-10 14:00 和 group_id，我想计算该 group_id 的事件数> 从 25-10 00:00 到 25-10 14:00。
做如下的事情:
df.groupBy('group_id', window('event_time', '1 hour').alias('model_window')) \
.agg(dfcount(lit(1)).alias('values'))
计算每小时的事件数，但不是每天累计。
有什么想法吗？
编辑: 预期的输出类似于:
+---------+---------------------------------------------+-------+
|group_id |model_window |values |
+---------+---------------------------------------------+-------+
|XXXX |[2017-10-25 00:00:00.0,2017-10-25 01:00:00.0]| 10 |
|XXXX |[2017-10-25 00:00:00.0,2017-10-25 02:00:00.0]| 17 |
|XXXX |[2017-10-25 00:00:00.0,2017-10-25 03:00:00.0]| 22 |
|YYYY |[2017-10-25 00:00:00.0,2017-10-25 01:00:00.0]| 0 |
|YYYY |[2017-10-25 00:00:00.0,2017-10-25 02:00:00.0]| 1 |
|YYYY |[2017-10-25 00:00:00.0,2017-10-25 03:00:00.0]| 9 |
+---------+---------------------------------------------+-------+
请您参考如下方法:
want to calculate ... per hour within a day per group_id.
提取数据和时间:
from pyspark.sql.functions import col, count, hour, sum
extended = (df
.withColumn("event_time", col("event_time").cast("timestamp"))
.withColumn("date", col("event_time").cast("date"))
.withColumn("hour", hour(col("event_time"))))
计算聚合
aggs = extended.groupBy("group_id", "date", "hour").count()
I want to calculate the rolling count of events
并使用窗口函数:
from pyspark.sql.window import Window
aggs.withColumn(
"agg_count",
sum("count").over(Window.partitionBy("group_id", "date").orderBy("hour")))
要获得 0 的缺失间隔，您必须为每个日期和小时生成引用数据并加入它。
df 定义为:
df = sc.parallelize([
("XXXX", "2017-10-25 01:47:02.717013"),
("XXXX", "2017-10-25 14:47:25.444979"),
("XXXX", "2017-10-25 14:49:32.21353"),
("YYYY", "2017-10-25 14:50:38.321134"),
("YYYY", "2017-10-25 14:51:12.028447"),
("ZZZZ", "2017-10-25 14:51:24.810688"),
("YYYY", "2017-10-25 14:37:34.241097"),
("ZZZZ", "2017-10-25 14:37:24.427836"),
("XXXX", "2017-10-25 22:37:24.620864"),
("YYYY", "2017-10-25 16:37:24.964614")
]).toDF(["group_id", "event_time"])
结果是
+--------+----------+----+-----+---------+
|group_id| date|hour|count|agg_count|
+--------+----------+----+-----+---------+
| XXXX|2017-10-25| 1| 1| 1|
| XXXX|2017-10-25| 14| 2| 3|
| XXXX|2017-10-25| 22| 1| 4|
| ZZZZ|2017-10-25| 14| 2| 2|
| YYYY|2017-10-25| 14| 3| 3|
| YYYY|2017-10-25| 16| 1| 4|
+--------+----------+----+-----+---------+
