Question

I am searching for a solution how to use the aggregate function to sum up a column given several criteria in other columns. R should select a range in a column and executean operation in the same rows considering the value from another row.

The practical problem I am trying to solve is following:
I got a list of electricity load measured every 15 minutes of a the day for every day over 2 years. It looks like this:

Date ______Time ______ Load
01-01-2010 00:00-00:15 1234

01-01-2010 00:15-00:30 2313

01-01-2010 …

01-01-2010 23:30-23:45 2341

…

31-12-2011 23:30-23:45 2347

My aim is to compute the so called “Peak-Load” and the “Off-Peak-Load”
The Peak is from 8 am to 8 pm. Off-Peak is the Opposite.
So I want to calculate the Peak and Off-Peak for every day.
Hence, I need aggregate for every day 8:00 to 20:00 and calculate the remaining load of the day.

I am also hap

Thanks for your help!

best,
F

Answer 1

I think your mental model of a heirarchy here is making this way too complicated. You don’t have to subset by day and then by peak/off-peak. Just subset jointly.

Using ddply:

dat <- data.frame(date=rep(seq(5),5),time=runif(25),load=rnorm(25))
library(plyr)
dat$peak <- dat$time<.5
ddply(dat, .(date,peak), function(x) mean(x$load) )

> ddply(dat, .(date,peak), function(x) mean(x$load) )
   date  peak           V1
1     1 FALSE -1.064166845
2     1  TRUE  0.172868201
3     2 FALSE  0.638594830
4     2  TRUE  0.045538051
5     3 FALSE  0.201264770
6     3  TRUE  0.054019462
7     4 FALSE  0.722268759
8     4  TRUE -0.490305933
9     5 FALSE  0.003411591
10    5  TRUE  0.628566966

Using aggregate:

> aggregate(dat$load, list(dat$date,dat$peak), mean )
   Group.1 Group.2            x
1        1   FALSE -1.064166845
2        2   FALSE  0.638594830
3        3   FALSE  0.201264770
4        4   FALSE  0.722268759
5        5   FALSE  0.003411591
6        1    TRUE  0.172868201
7        2    TRUE  0.045538051
8        3    TRUE  0.054019462
9        4    TRUE -0.490305933
10       5    TRUE  0.628566966

And just for the fun of it, benchmarks

First, using 5×5 entries as above:

> microbenchmark(
+   ddply(dat, .(date,peak), function(x) mean(x$load) ),
+   aggregate(dat$load, list(dat$date,dat$peak), mean )
+   )
Unit: milliseconds
                                                 expr      min       lq   median       uq      max
1 aggregate(dat$load, list(dat$date, dat$peak), mean) 1.323438 1.376635 1.445769 1.549663 2.853348
2 ddply(dat, .(date, peak), function(x) mean(x$load)) 4.057177 4.292442 4.386289 4.534728 6.864962

Next using 500×500 entries

> m
Unit: milliseconds
                                                 expr      min       lq   median       uq      max
1 aggregate(dat$load, list(dat$date, dat$peak), mean) 558.9524 570.7354 590.4633 599.4404 634.3201
2 ddply(dat, .(date, peak), function(x) mean(x$load)) 317.7781 348.1116 361.7118 413.4490 503.8540

50×50 benchmarks

n <- 50
dat <- data.frame(date=rep(seq(n),n),time=runif(n),load=rnorm(n))
dat$peak <- dat$time<.5

library(plyr)
library(microbenchmark)
library(data.table)
DT <- as.data.table(dat)
m <- microbenchmark(
  ddply(dat, .(date,peak), function(x) mean(x$load) ),
  aggregate(dat$load, list(dat$date,dat$peak), mean ),
  DT[,.Internal(mean(load)),keyby=list(date,peak)]
  )
m
plot(m)

So aggregate is faster for small problems (presumably because it has less overhead to load up all the machinery), and ddply is faster for large problems (where speed matters). Data.table blows everything away (as usual).