Aggregate multiple columns based on half-hourly time series

Question

I have a timeseries with a 5 minute resolution and I would like to aggregate (average) all values within 30 minutes intervals for each category (X/Y/Z) variable.

I have found many threads here on SO but none of them aggregates half-hourly and I don't have an idea how to combine the half-hourly aggregation with a second column. Furthermore, I would appreciate it if there would be a data.table solution, because of the superior performance. The original tables are several million rows long with 1000s of categories.

My data looks like this:

+---------------------+------+------------+
|      Timestamp      | DUID | Meter_Prod |
+---------------------+------+------------+
| 2018-03-01 00:00:00 | X    |          1 |
| 2018-03-01 00:00:00 | Y    |          2 |
| 2018-03-01 00:00:00 | Z    |          3 |
| 2018-03-01 00:05:00 | X    |          1 |
| 2018-03-01 00:05:00 | Y    |          2 |
| 2018-03-01 00:05:00 | Z    |          3 |
| ...                 |      |            |
| 2018-03-01 00:55:00 | X    |          1 |
| 2018-03-01 00:55:00 | Y    |          2 |
| 2018-03-01 00:55:00 | Z    |          3 |
+---------------------+------+------------+

I would like to have this

+---------------------+------+--------------------+
|      Timestamp      | DUID | Meter_Prod_Average |
+---------------------+------+--------------------+
| 2018-03-01 00:00:00 | X    |                  1 |
| 2018-03-01 00:00:00 | Y    |                  2 |
| 2018-03-01 00:00:00 | Z    |                  3 |
| 2018-03-01 00:30:00 | X    |                  1 |
| 2018-03-01 00:30:00 | Y    |                  2 |
| 2018-03-01 00:30:00 | Z    |                  3 |
+---------------------+------+--------------------+

An example data frame is uploaded here: https://pastebin.com/4bESGTKH

There are some posts related to that stackoverflow.com/questions/39987875/… , stackoverflow.com/questions/40252525/… and stackoverflow.com/questions/27594959/… — Ronak Shah

Rohit Rohit · Accepted Answer · 2019-03-12T10:38:09

You could try a data.table rolling join. Create a new data.table with the time intervals you want:

head(dt)
# Timestamp DUID Meter_Prod
# 1 2018-03-01 00:00:00    X         15
# 2 2018-03-01 00:00:00    Y        122
# 3 2018-03-01 00:00:00    Z          6
# 4 2018-03-01 00:05:00    X         15
# 5 2018-03-01 00:05:00    Y        122
# 6 2018-03-01 00:05:00    Z          6
start_time <- as.POSIXct('2018-03-01 00:00:00')
new_time <- data.table(Timestamp=seq.POSIXt(start_time,max(dt$Timestamp),by=30*60)) # Will be deleted during join
new_time[,ts:=Timestamp] # Will be preserved for grouping

The new table should have a dummy column with the same timestamps for the join.

new_dt <- new_time[dt,on='Timestamp',roll=+Inf] #Join statement
new_dt[,.(Meter_Prod_Average=mean(Meter_Prod)),by=.(ts,DUID)] # Aggregation
#                      ts DUID Meter_Prod_Average
# 1: 2018-03-01 00:00:00    X                 15
# 2: 2018-03-01 00:00:00    Y                122
# 3: 2018-03-01 00:00:00    Z                  6
# 4: 2018-03-01 00:30:00    X                 15
# 5: 2018-03-01 00:30:00    Y                122
# ---                                            
#   140: 2018-03-01 23:00:00    Y                122
# 141: 2018-03-01 23:00:00    Z                  6
# 142: 2018-03-01 23:30:00    X               2696
# 143: 2018-03-01 23:30:00    Y                122
# 144: 2018-03-01 23:30:00    Z                  6

Based on the roll argument in the join you can vary the position of the timestamp wrt the interval.

Aggregate multiple columns based on half-hourly time series

1 Answers