Dear All,
I am trying to propagate the last valid observation (e.g. not null) to the null
values in a dataset.
Below I reported the partial solution:
Dataset tmp800=tmp700.select("uuid", "eventTime", "Washer_rinseCycles");
WindowSpec wspec=
Window.partitionBy(tmp800.col("uuid")).orderBy(tmp800.col("uuid"),tmp800.col("eventTime"));
Column c1 =
org.apache.spark.sql.functions.lag(tmp800.col("Washer_rinseCycles"),1).over(wspec);
Dataset tmp900=tmp800.withColumn("Washer_rinseCyclesFilled",
when(tmp800.col("Washer_rinseCycles").isNull(),
c1).otherwise(tmp800.col("Washer_rinseCycles")));
However, It does not solve the entire problem as the function lag(,1) returns
the value that is the rows before the current row even if it is NULL (see the
below table).
Is there in SPARK a similar method to Pandas' "backfill" for the DataFrame?
Is it possible to do it using SPARK API? How?
Many Thanks in advance.
Best Regards,
Carlo
[Immagine in linea con il testo]