Join two dataframes using the closest timestamp pyspark

Question

So I am very new to pyspark but I am still unable to correctly create my own query. I try googling my problems but I just don't understand how most of this works. I'm not sure what I'm missing.

But anyway I have the following two dataframes, spark_p:

+--------------------+-----+--------------------+
|__record_timestamp__|cycle|            profiles|
+--------------------+-----+--------------------+
| 1651737406300000000|    0|[0, 1, 1, 1, 3, 1...|
| 1651737406300000000|   16|[0, 0, 1, 0, 0, 0...|
| 1651737406300000000|   17|[1, 1, 1, 1, 0, 0...|
| 1651737406300000000|   18|[0, 0, 0, 0, 0, 1...|
| 1651737406300000000|   19|[1, 1, 1, 0, 0, 0...|
+--------------------+-----+--------------------+

and spark_m:

+-------------+--------------------+
|      current|__record_timestamp__|
+-------------+--------------------+
|  0.007181627| 1651730407500000000|
| 8.3004625E-4| 1651730464000000000|
|   0.41976404| 1651730507000000000|
|-0.0017322368| 1651732761000000000|
|-2.5260705E-4| 1651732822500000000|
| 2.3460487E-4| 1651732824500000000|
+-------------+--------------------+

And I need to add a column to spark_p that contains the current at that specific timestamp.

So the result would look something like:

+--------------------+-----+--------------------+---------+
|__record_timestamp__|cycle|            profiles|  current|
+--------------------+-----+--------------------+---------+
| 1651737406300000000|    0|[0, 1, 1, 1, 3, 1...|     0.07|
| 1651737406300000000|   16|[0, 0, 1, 0, 0, 0...|       12|
| 1651737406300000000|   17|[1, 1, 1, 1, 0, 0...|      0.0|
| 1651737406300000000|   18|[0, 0, 0, 0, 0, 1...| 5.235654|
| 1651737406300000000|   19|[1, 1, 1, 0, 0, 0...|      125|
+--------------------+-----+--------------------+---------+

Now the time stamps won't exactly match up but I just need the closest timestamp, or to use the value of the previously recorded current, either is fine. I have no idea how...

When I try:

spark_p.join(spark_m, spark_p.__record_timestamp__ ==  spark_m.__record_timestamp__, "inner").show()

I just get:

+--------------------+-----+--------+-----+--------------------+
|__record_timestamp__|cycle|profiles|value|__record_timestamp__|
+--------------------+-----+--------+-----+--------------------+
+--------------------+-----+--------+-----+--------------------+

So I'm guessing none of them match exactly, but how would I just grab the nearest value? TIA

Answer 1

This solution contains the answer:

SPLIT_COUNT = 90
SPLIT_SIZE = 1024

spark_p = data.select("profiles", '__record_timestamp__')
spark_p = spark_p.withColumn("profiles", F.col("profiles").getField("elements") )

slices = [F.slice(F.col('profiles'), i * SPLIT_SIZE + 1, SPLIT_SIZE) for i in range(SPLIT_COUNT)]

spark_p = spark_p.select(F.posexplode(F.array(*slices)), F.col('__record_timestamp__'))
spark_p = spark_p.withColumn("cycle", F.col("pos") )
spark_p = spark_p.withColumn("profiles", F.col("col") )
spark_p = spark_p.drop('pos').drop('col')

spark_m = magnetData.select("value", '__record_timestamp__', )


spark_p = spark_p.withColumn('value', F.lit(None))


spark_m = spark_m.withColumn('profiles', F.lit(None))
spark_m = spark_m.withColumn('cycle', F.lit(None))


final_df = spark_p.unionByName(spark_m)

w = Window.orderBy('__record_timestamp__').rowsBetween(Window.unboundedPreceding, -1)

final_df = final_df.withColumn('value', F.last('value', True).over(w)).filter(~F.isnull('profiles'))

You must create a window with unboundedPreceding parameter.