CODEWITHSUNDEEP
apache-sparkpysparkapache-spark-sql
Into Numbers
Into Numbers

Reputation: 963

Join two timeseries dataframes to get the most recent right entry for each left entry in PySpark

I have two Sparks dataframes:

df1 with one entry per id and date:

|date       |id   |
+-----------+-----+
|2021-11-15 |    1|
|2021-11-14 |    1|
|2021-11-15 |    2|
|2021-11-14 |    2|
|2021-11-15 |    3|
|2021-11-14 |    3|

df2 with multiple log entries:

|date       |id   |
+-----------+-----+
|2021-11-13 |    1|
|2021-11-13 |    1|
|2021-11-13 |    3|
|2021-11-14 |    1|
|2021-11-14 |    1|
|2021-11-14 |    1|
|2021-11-14 |    1|
|2021-11-15 |    1|
|2021-11-15 |    1|

how can I join these dfs, so that I get the most recent possible entry (date(df2) should be <= date) per id and date of df2?

|date       |id    |   date(df2)| 
+-----------+------+------------+
|2021-11-15 |    1 | 2021-11-15 |
|2021-11-14 |    1 | 2021-11-14 |
|2021-11-15 |    2 |       null |
|2021-11-14 |    2 |       null |
|2021-11-15 |    3 | 2021-11-13 |
|2021-11-14 |    3 | 2021-11-13 |

THX Into Numbers

Upvotes: 0

Views: 265

Answers (1)

blackbishop
blackbishop

Reputation: 32650

Use join then group by df1.id and df2.date and use conditional aggregation to get max df2.date <= df1.date

import pyspark.sql.functions as F


result_df = df1.join(
    df2.withColumnRenamed("date", "df2_date"),
    ["id"],
    "left"
).groupBy("id", "date").agg(
    F.max(
        F.when(F.col("df2_date") <= F.col("date"), F.col("df2_date"))
    ).alias("df2_date")
)

result_df.show()
#+---+----------+----------+
#| id|      date|  df2_date|
#+---+----------+----------+
#|  1|2021-11-14|2021-11-14|
#|  1|2021-11-15|2021-11-15|
#|  2|2021-11-14|      null|
#|  2|2021-11-15|      null|
#|  3|2021-11-14|2021-11-13|
#|  3|2021-11-15|2021-11-13|
#+---+----------+----------+

Upvotes: 1

Related Questions